Pro
EN
Sign inAccount Download for Mac
Download for Mac
Mac requirements M1 · M2 · M3 · M4 · M5

Will my Mac run it?

Phosphene runs on any Apple Silicon Mac, M1 to M5, with 16 GB of memory or more: MacBook Air, MacBook Pro, Mac mini, iMac and Mac Studio. Memory decides what you can make. At 16 GB you get LTX-2.5 video with sound from text or a picture, MiniMax H3 video with dialogue, FLUX.2 pictures, songs and speech. 24 GB adds character training and picture editing. 36 GB adds long H3 clips and MiniMax-H3-Image. 48 GB adds full-size LTX-2.5, High, keyframes and Extend. Intel Macs and Windows are not supported.

Find your memory: Apple menu → About This Mac → Memory.

Phosphene reads your memory when it starts and only offers what fits.

Facts from Phosphene 5.0.26. Render times from 2,279 Macs, Aug 17, 2026 to Nov 15, 2026.

Pick your Mac's memory

Or pick your chip

8 GB

Not supported. Some short LTX-2.5 clips do finish, slowly.

Below 16 GB Phosphene is not supported.

A 5-second clip with sound took 9 min 10 s on M2 to 13 min 20 s on M1 with 8 GB (median, model loading included). 480p

Macs with 8 GB in our render data: M1, M2.

Will it run on my 8 GB Mac?

16 GB

Video with sound, H3 dialogue, pictures, songs and speech.

With 16 GB: 6 of 29 things Phosphene makes.

A 5-second clip with sound took 3 min 55 s on M5 to 13 min 20 s on M1 with 16 GB (median, model loading included). 480p

Macs with 16 GB in our render data: M1, M1 Pro, M2, M2 Pro, M3, M4, M5.

Video: LTX-2.52/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. New at 16 GB
  • Full-size LTX-2.5 video Needs 48 GB
  • High quality Needs 48 GB
  • Keyframes, start and end frame, Extend Needs 48 GB
  • Render your trained characters Needs 48 GB
  • High · 720p Needs 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. New at 16 GB
  • Pro: AI fill for Make it vertical Needs 48 GB

Video with dialogue: MiniMax H31/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. New at 16 GB
  • MiniMax H3 at High (1024×576) Needs 18 GB
  • Long H3 clips: 10 and 15 seconds Needs 36 GB
  • H3 Clean up sound Needs 36 GB
  • H3's full bf16 engine Needs 64 GB
  • Pro: faster H3 picture to video Needs 36 GB
  • Pro: faster H3 Flex Turbo Needs 36 GB

Pictures1/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. New at 16 GB
  • Picture editing: Reference Edit (Qwen-Image-Edit) Needs 24 GB
  • FLUX 3 Klein at Q4 Needs 24 GB
  • Ideogram 4 (words on the picture), Fast mode Needs 24 GB
  • FLUX 3 Klein at Q6 and Q8 Needs 32 GB
  • Ideogram 4 at its default quality Needs 32 GB
  • P-H3-Image: large photographs and posters from text Needs 32 GB
  • MiniMax-H3-Image: edit and combine up to 6 pictures Needs 36 GB

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. New at 16 GB
  • Speech from a script Nine preset voices. New at 16 GB

Training0/2

  • Train an original character, lighter recipe Needs 24 GB
  • Train an original character, full recipe Needs 64 GB

18 GB

Adds MiniMax H3 at High size.

With 18 GB: 7 of 29 things Phosphene makes.

A 5-second clip with sound took 4 min 35 s on M3 Pro with 18 GB (median, model loading included). 480p

Macs with 18 GB in our render data: M3 Pro.

Video: LTX-2.52/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Needs 48 GB
  • High quality Needs 48 GB
  • Keyframes, start and end frame, Extend Needs 48 GB
  • Render your trained characters Needs 48 GB
  • High · 720p Needs 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Needs 48 GB

Video with dialogue: MiniMax H32/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. New at 18 GB
  • Long H3 clips: 10 and 15 seconds Needs 36 GB
  • H3 Clean up sound Needs 36 GB
  • H3's full bf16 engine Needs 64 GB
  • Pro: faster H3 picture to video Needs 36 GB
  • Pro: faster H3 Flex Turbo Needs 36 GB

Pictures1/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Needs 24 GB
  • FLUX 3 Klein at Q4 Needs 24 GB
  • Ideogram 4 (words on the picture), Fast mode Needs 24 GB
  • FLUX 3 Klein at Q6 and Q8 Needs 32 GB
  • Ideogram 4 at its default quality Needs 32 GB
  • P-H3-Image: large photographs and posters from text Needs 32 GB
  • MiniMax-H3-Image: edit and combine up to 6 pictures Needs 36 GB

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training0/2

  • Train an original character, lighter recipe Needs 24 GB
  • Train an original character, full recipe Needs 64 GB

24 GB

Adds character training, picture editing and more picture models.

With 24 GB: 11 of 29 things Phosphene makes.

A 5-second clip with sound took 1 min 35 s on M5 Pro to 7 min 50 s on M4 with 24 GB (median, model loading included). 480p

Macs with 24 GB in our render data: M2, M3, M4, M4 Pro, M5, M5 Pro.

Video: LTX-2.52/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Needs 48 GB
  • High quality Needs 48 GB
  • Keyframes, start and end frame, Extend Needs 48 GB
  • Render your trained characters Needs 48 GB
  • High · 720p Needs 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Needs 48 GB

Video with dialogue: MiniMax H32/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Needs 36 GB
  • H3 Clean up sound Needs 36 GB
  • H3's full bf16 engine Needs 64 GB
  • Pro: faster H3 picture to video Needs 36 GB
  • Pro: faster H3 Flex Turbo Needs 36 GB

Pictures4/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. New at 24 GB
  • FLUX 3 Klein at Q4 A small picture and one reference. New at 24 GB
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. New at 24 GB
  • FLUX 3 Klein at Q6 and Q8 Needs 32 GB
  • Ideogram 4 at its default quality Needs 32 GB
  • P-H3-Image: large photographs and posters from text Needs 32 GB
  • MiniMax-H3-Image: edit and combine up to 6 pictures Needs 36 GB

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training1/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. New at 24 GB
  • Train an original character, full recipe Needs 64 GB

32 GB

Adds the larger picture models.

With 32 GB: 14 of 29 things Phosphene makes.

A 5-second clip with sound took 2 min 45 s on M2 Max to 7 min on M4 with 32 GB (median, model loading included). 480p

Macs with 32 GB in our render data: M1 Pro, M1 Max, M2 Pro, M2 Max, M4, M5.

Video: LTX-2.52/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Needs 48 GB
  • High quality Needs 48 GB
  • Keyframes, start and end frame, Extend Needs 48 GB
  • Render your trained characters Needs 48 GB
  • High · 720p Needs 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Needs 48 GB

Video with dialogue: MiniMax H32/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Needs 36 GB
  • H3 Clean up sound Needs 36 GB
  • H3's full bf16 engine Needs 64 GB
  • Pro: faster H3 picture to video Needs 36 GB
  • Pro: faster H3 Flex Turbo Needs 36 GB

Pictures7/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. Runs
  • FLUX 3 Klein at Q4 A small picture and one reference. Runs
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. Runs
  • FLUX 3 Klein at Q6 and Q8 Q8 is the recommended quality. New at 32 GB
  • Ideogram 4 at its default quality New at 32 GB
  • P-H3-Image: large photographs and posters from text Needs about 26 GB of free memory and MiniMax H3 installed. New at 32 GB
  • MiniMax-H3-Image: edit and combine up to 6 pictures Needs 36 GB

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training1/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. Runs
  • Train an original character, full recipe Needs 64 GB

36 GB

Adds long H3 clips, Clean up sound and MiniMax-H3-Image. H3 keeps its engine in memory from here.

With 36 GB: 19 of 29 things Phosphene makes.

A 5-second clip with sound took 55 s on M5 Max to 2 min 40 s on M3 Pro with 36 GB (median, model loading included). 480p

Macs with 36 GB in our render data: M3 Pro, M3 Max, M4 Max, M5 Max.

Video: LTX-2.52/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Needs 48 GB
  • High quality Needs 48 GB
  • Keyframes, start and end frame, Extend Needs 48 GB
  • Render your trained characters Needs 48 GB
  • High · 720p Needs 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Needs 48 GB

Video with dialogue: MiniMax H36/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Joined 5-second sections that continue the movement and sound. New at 36 GB
  • H3 Clean up sound New at 36 GB
  • H3's full bf16 engine Needs 64 GB
  • Pro: faster H3 picture to video A 5-second clip took 10:02 on Pro against 14:57 on Free (M4 Max, 64 GB). In the Mac app. New at 36 GB
  • Pro: faster H3 Flex Turbo About 1.5 times faster than H3 Flex on Free (M4 Max, 64 GB). In the Pinokio install. New at 36 GB

Pictures8/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. Runs
  • FLUX 3 Klein at Q4 A small picture and one reference. Runs
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. Runs
  • FLUX 3 Klein at Q6 and Q8 Q8 is the recommended quality. Runs
  • Ideogram 4 at its default quality Runs
  • P-H3-Image: large photographs and posters from text Needs about 26 GB of free memory and MiniMax H3 installed. Runs
  • MiniMax-H3-Image: edit and combine up to 6 pictures Also text to image at 4, 8 and 16 megapixels. Needs MiniMax H3 installed. Editing one picture took about 5 minutes on an M4 Max with 64 GB. New at 36 GB

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training1/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. Runs
  • Train an original character, full recipe Needs 64 GB

48 GB

Adds full-size LTX-2.5, High, keyframes, Extend and your trained characters.

With 48 GB: 24 of 29 things Phosphene makes.

A 5-second clip with sound took 1 min 20 s on M5 Max to 5 min 50 s on M4 Pro with 48 GB (median, model loading included). 576p

Macs with 48 GB in our render data: M3 Max, M4 Pro, M4 Max, M5 Pro, M5 Max.

Video: LTX-2.57/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Balanced at 1024×576 and Standard at 1280×704. New at 48 GB
  • High quality A second, larger pass for sharper detail and steadier motion. Needs the High add-on (29.5 GB). New at 48 GB
  • Keyframes, start and end frame, Extend Capped at 768 px on the long side from 48 to 79 GB. New at 48 GB
  • Render your trained characters Needs the Q8 weights (30.02 GB). Q8 Draft is comfortable at 64 GB, Q8 Pro at 96 GB. New at 48 GB
  • High · 720p Needs 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Paints real picture above and below. About 15 minutes for a 5-second clip. New at 48 GB

Video with dialogue: MiniMax H36/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Joined 5-second sections that continue the movement and sound. Runs
  • H3 Clean up sound Runs
  • H3's full bf16 engine Needs 64 GB
  • Pro: faster H3 picture to video A 5-second clip took 10:02 on Pro against 14:57 on Free (M4 Max, 64 GB). In the Mac app. Runs
  • Pro: faster H3 Flex Turbo About 1.5 times faster than H3 Flex on Free (M4 Max, 64 GB). In the Pinokio install. Runs

Pictures8/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. Runs
  • FLUX 3 Klein at Q4 A small picture and one reference. Runs
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. Runs
  • FLUX 3 Klein at Q6 and Q8 Q8 is the recommended quality. Runs
  • Ideogram 4 at its default quality Runs
  • P-H3-Image: large photographs and posters from text Needs about 26 GB of free memory and MiniMax H3 installed. Runs
  • MiniMax-H3-Image: edit and combine up to 6 pictures Also text to image at 4, 8 and 16 megapixels. Needs MiniMax H3 installed. Editing one picture took about 5 minutes on an M4 Max with 64 GB. Runs

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training1/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. Runs
  • Train an original character, full recipe Needs 64 GB

64 GB

Adds High · 720p, the full character training recipe and H3's full engine.

With 64 GB: 27 of 29 things Phosphene makes.

A 5-second clip with sound took 1 min 20 s on M5 Max to 5 min 40 s on M1 Max with 64 GB (median, model loading included). 576p

Macs with 64 GB in our render data: M1 Max, M1 Ultra, M2 Max, M2 Ultra, M3 Max, M4 Pro, M4 Max, M5 Pro, M5 Max.

Video: LTX-2.58/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Balanced at 1024×576 and Standard at 1280×704. Runs
  • High quality A second, larger pass for sharper detail and steadier motion. Needs the High add-on (29.5 GB). Runs
  • Keyframes, start and end frame, Extend Capped at 768 px on the long side from 48 to 79 GB. Runs
  • Render your trained characters Needs the Q8 weights (30.02 GB). Q8 Draft is comfortable at 64 GB, Q8 Pro at 96 GB. Runs
  • High · 720p 1280×704, the most detail LTX-2.5 makes. Measured peak 49.7 GB. New at 64 GB
  • Keyframes and Extend up to 1024 px Needs 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Paints real picture above and below. About 15 minutes for a 5-second clip. Runs

Video with dialogue: MiniMax H37/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Joined 5-second sections that continue the movement and sound. Runs
  • H3 Clean up sound Runs
  • H3's full bf16 engine From 60 GB. Below that, H3 runs its compact Q8 engine, built on your Mac once. New at 64 GB
  • Pro: faster H3 picture to video A 5-second clip took 10:02 on Pro against 14:57 on Free (M4 Max, 64 GB). In the Mac app. Runs
  • Pro: faster H3 Flex Turbo About 1.5 times faster than H3 Flex on Free (M4 Max, 64 GB). In the Pinokio install. Runs

Pictures8/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. Runs
  • FLUX 3 Klein at Q4 A small picture and one reference. Runs
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. Runs
  • FLUX 3 Klein at Q6 and Q8 Q8 is the recommended quality. Runs
  • Ideogram 4 at its default quality Runs
  • P-H3-Image: large photographs and posters from text Needs about 26 GB of free memory and MiniMax H3 installed. Runs
  • MiniMax-H3-Image: edit and combine up to 6 pictures Also text to image at 4, 8 and 16 megapixels. Needs MiniMax H3 installed. Editing one picture took about 5 minutes on an M4 Max with 64 GB. Runs

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training2/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. Runs
  • Train an original character, full recipe The High preset, measured for character consistency. New at 64 GB

96 GB

Keyframes and Extend go up to 1024 px.

With 96 GB: 28 of 29 things Phosphene makes.

A 5-second clip with sound took 1 min 50 s on M3 Ultra to 3 min 35 s on M2 Max with 96 GB (median, model loading included). 576p

Macs with 96 GB in our render data: M2 Max, M3 Ultra, M5 Ultra.

Video: LTX-2.59/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Balanced at 1024×576 and Standard at 1280×704. Runs
  • High quality A second, larger pass for sharper detail and steadier motion. Needs the High add-on (29.5 GB). Runs
  • Keyframes, start and end frame, Extend Capped at 768 px on the long side from 48 to 79 GB. Runs
  • Render your trained characters Needs the Q8 weights (30.02 GB). Q8 Draft is comfortable at 64 GB, Q8 Pro at 96 GB. Runs
  • High · 720p 1280×704, the most detail LTX-2.5 makes. Measured peak 49.7 GB. Runs
  • Keyframes and Extend up to 1024 px From 80 GB. New at 96 GB
  • No size limits Needs 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Paints real picture above and below. About 15 minutes for a 5-second clip. Runs

Video with dialogue: MiniMax H37/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Joined 5-second sections that continue the movement and sound. Runs
  • H3 Clean up sound Runs
  • H3's full bf16 engine From 60 GB. Below that, H3 runs its compact Q8 engine, built on your Mac once. Runs
  • Pro: faster H3 picture to video A 5-second clip took 10:02 on Pro against 14:57 on Free (M4 Max, 64 GB). In the Mac app. Runs
  • Pro: faster H3 Flex Turbo About 1.5 times faster than H3 Flex on Free (M4 Max, 64 GB). In the Pinokio install. Runs

Pictures8/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. Runs
  • FLUX 3 Klein at Q4 A small picture and one reference. Runs
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. Runs
  • FLUX 3 Klein at Q6 and Q8 Q8 is the recommended quality. Runs
  • Ideogram 4 at its default quality Runs
  • P-H3-Image: large photographs and posters from text Needs about 26 GB of free memory and MiniMax H3 installed. Runs
  • MiniMax-H3-Image: edit and combine up to 6 pictures Also text to image at 4, 8 and 16 megapixels. Needs MiniMax H3 installed. Editing one picture took about 5 minutes on an M4 Max with 64 GB. Runs

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training2/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. Runs
  • Train an original character, full recipe The High preset, measured for character consistency. Runs

128 GB or more

No size limits on any LTX-2.5 mode.

With 128 GB or more: everything Phosphene makes.

A 5-second clip with sound took 40 s on M5 Ultra to 2 min 55 s on M4 Max with 128+ GB (median, model loading included). 576p

Macs with 128+ GB in our render data: M1 Ultra, M2 Ultra, M3 Max, M3 Ultra, M4 Max, M5 Max, M5 Ultra.

Video: LTX-2.510/10

  • Text to video and picture to video, with sound Up to 768 px on the long side (for example 768×432), Quick and Balanced, 3 to 10 seconds. Macs with 24 GB or less use a low-memory path automatically. On a 16 GB Mac a clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. Runs
  • Full-size LTX-2.5 video Balanced at 1024×576 and Standard at 1280×704. Runs
  • High quality A second, larger pass for sharper detail and steadier motion. Needs the High add-on (29.5 GB). Runs
  • Keyframes, start and end frame, Extend Capped at 768 px on the long side from 48 to 79 GB. Runs
  • Render your trained characters Needs the Q8 weights (30.02 GB). Q8 Draft is comfortable at 64 GB, Q8 Pro at 96 GB. Runs
  • High · 720p 1280×704, the most detail LTX-2.5 makes. Measured peak 49.7 GB. Runs
  • Keyframes and Extend up to 1024 px From 80 GB. Runs
  • No size limits From 120 GB, every LTX-2.5 mode at any size the model supports. New at 128 GB
  • Pro: faster LTX-2.5 clips Every Mac from 16 GB. About 27 to 30% faster than Free on 16 to 24 GB Macs; about a quarter less waiting on a 5-second Balanced clip on an M4 Max with 64 GB. Runs
  • Pro: AI fill for Make it vertical Paints real picture above and below. About 15 minutes for a 5-second clip. Runs

Video with dialogue: MiniMax H37/7

  • MiniMax H3: 3 and 5-second clips at Draft and Standard Video, dialogue and sound from one prompt. Below 36 GB it reads the model from your drive as it renders; the clip comes out the same. Tested on a 64 GB Mac limited to 16 GB; no render times from 16 GB Macs yet. Runs
  • MiniMax H3 at High (1024×576) Sharper faces than Standard. Runs
  • Long H3 clips: 10 and 15 seconds Joined 5-second sections that continue the movement and sound. Runs
  • H3 Clean up sound Runs
  • H3's full bf16 engine From 60 GB. Below that, H3 runs its compact Q8 engine, built on your Mac once. Runs
  • Pro: faster H3 picture to video A 5-second clip took 10:02 on Pro against 14:57 on Free (M4 Max, 64 GB). In the Mac app. Runs
  • Pro: faster H3 Flex Turbo About 1.5 times faster than H3 Flex on Free (M4 Max, 64 GB). In the Pinokio install. Runs

Pictures8/8

  • Pictures from text: FLUX.2 [klein] 4B What Auto uses on a 16 GB Mac. Runs
  • Picture editing: Reference Edit (Qwen-Image-Edit) Its smallest size from 24 GB; 32 GB at 6-bit, 36 GB at 8-bit. Runs
  • FLUX 3 Klein at Q4 A small picture and one reference. Runs
  • Ideogram 4 (words on the picture), Fast mode Non-commercial licence. Runs
  • FLUX 3 Klein at Q6 and Q8 Q8 is the recommended quality. Runs
  • Ideogram 4 at its default quality Runs
  • P-H3-Image: large photographs and posters from text Needs about 26 GB of free memory and MiniMax H3 installed. Runs
  • MiniMax-H3-Image: edit and combine up to 6 pictures Also text to image at 4, 8 and 16 megapixels. Needs MiniMax H3 installed. Editing one picture took about 5 minutes on an M4 Max with 64 GB. Runs

Songs and speech2/2

  • Songs with vocals (YuE2 Music Studio) Peaks around 11 GB. 24 GB or more is better for long songs with other apps open. Runs
  • Speech from a script Nine preset voices. Runs

Training2/2

  • Train an original character, lighter recipe Fewer steps and a lower rank so it finishes. Trains on LTX-2.3. Runs
  • Train an original character, full recipe The High preset, measured for character consistency. Runs
02 / 07

Everything, by memory

One row per thing Phosphene makes, one column per memory size. A tick means the app offers it on a Mac with that much memory.

What Phosphene 5.0.26 runs at each memory size
What you make 8 GB16 GB18 GB24 GB32 GB36 GB48 GB64 GB96 GB128 GB or more
Video: LTX-2.5
Text to video and picture to video, with sound NoRunsRunsRunsRunsRunsRunsRunsRunsRuns
Full-size LTX-2.5 video NoNoNoNoNoNoRunsRunsRunsRuns
High quality NoNoNoNoNoNoRunsRunsRunsRuns
Keyframes, start and end frame, Extend NoNoNoNoNoNoRunsRunsRunsRuns
Render your trained characters NoNoNoNoNoNoRunsRunsRunsRuns
High · 720p NoNoNoNoNoNoNoRunsRunsRuns
Keyframes and Extend up to 1024 px NoNoNoNoNoNoNoNoRunsRuns
No size limits NoNoNoNoNoNoNoNoNoRuns
Pro: faster LTX-2.5 clips NoProProProProProProProProPro
Pro: AI fill for Make it vertical NoNoNoNoNoNoProProProPro
Video with dialogue: MiniMax H3
MiniMax H3: 3 and 5-second clips at Draft and Standard NoRunsRunsRunsRunsRunsRunsRunsRunsRuns
MiniMax H3 at High (1024×576) NoNoRunsRunsRunsRunsRunsRunsRunsRuns
Long H3 clips: 10 and 15 seconds NoNoNoNoNoRunsRunsRunsRunsRuns
H3 Clean up sound NoNoNoNoNoRunsRunsRunsRunsRuns
H3's full bf16 engine NoNoNoNoNoNoNoRunsRunsRuns
Pro: faster H3 picture to video NoNoNoNoNoProProProProPro
Pro: faster H3 Flex Turbo NoNoNoNoNoProProProProPro
Pictures
Pictures from text: FLUX.2 [klein] 4B NoRunsRunsRunsRunsRunsRunsRunsRunsRuns
Picture editing: Reference Edit (Qwen-Image-Edit) NoNoNoRunsRunsRunsRunsRunsRunsRuns
FLUX 3 Klein at Q4 NoNoNoRunsRunsRunsRunsRunsRunsRuns
Ideogram 4 (words on the picture), Fast mode NoNoNoRunsRunsRunsRunsRunsRunsRuns
FLUX 3 Klein at Q6 and Q8 NoNoNoNoRunsRunsRunsRunsRunsRuns
Ideogram 4 at its default quality NoNoNoNoRunsRunsRunsRunsRunsRuns
P-H3-Image: large photographs and posters from text NoNoNoNoRunsRunsRunsRunsRunsRuns
MiniMax-H3-Image: edit and combine up to 6 pictures NoNoNoNoNoRunsRunsRunsRunsRuns
Songs and speech
Songs with vocals (YuE2 Music Studio) NoRunsRunsRunsRunsRunsRunsRunsRunsRuns
Speech from a script NoRunsRunsRunsRunsRunsRunsRunsRunsRuns
Training
Train an original character, lighter recipe NoNoNoRunsRunsRunsRunsRunsRunsRuns
Train an original character, full recipe NoNoNoNoNoNoNoRunsRunsRuns
03 / 07

By chip: M1, M2, M3, M4 and M5

Memory decides what runs. The chip decides how fast. Within a generation, Pro and Max chips have more graphics cores and more memory bandwidth, and video models use both.

ChipFound inMemory seen in our render data
M1 to M5 MacBook Air, MacBook Pro 13-inch (M1, M2) and 14-inch (M3 and later), Mac mini, iMac
M1 Pro to M5 Pro MacBook Pro 14 and 16-inch, Mac mini (M2 Pro, M4 Pro)
M1 Max to M5 Max MacBook Pro 14 and 16-inch, Mac Studio
M1 Ultra to M3 Ultra Mac Studio, Mac Pro (M2 Ultra)
04 / 07

How long does a clip take on my Mac?

Median times across Phosphene users with the same chip and memory, from Aug 17, 2026 to Nov 15, 2026, model loading included. Thin rows come from three or four Macs; treat them as a guide.

Median render time by chip and memory
Chip Memory LTX-2.5, Balanced, 5 s LTX-2.5, Standard 720p, 5 s MiniMax H3, Standard, 5 s, Fast
M1 to M5
M1 8 GB 13 min 20 s480pnot measured yetnot measured yet
M1 16 GB 13 min 20 s480pnot measured yetnot measured yet
M2 8 GB 9 min 10 s480pnot measured yetnot measured yet
M2 16 GB 15 min480pfew Macsnot measured yetnot measured yet
M2 24 GB 8 min 40 s480pfew Macsnot measured yetnot measured yet
M3 16 GB 15 min 50 s480pfew Macsnot measured yetnot measured yet
M3 24 GB 23 min480pfew Macsnot measured yetnot measured yet
M4 16 GB 8 min 20 s480pnot measured yetnot measured yet
M4 24 GB 7 min 50 s480pnot measured yetnot measured yet
M4 32 GB 7 min480pnot measured yetnot measured yet
M5 16 GB 3 min 55 s480pnot measured yetnot measured yet
M5 24 GB 3 min 10 s480pnot measured yetnot measured yet
M5 32 GB 4 min 5 s480pnot measured yetnot measured yet
M1 Pro to M5 Pro
M1 Pro 16 GB 6 min 10 s480pnot measured yetnot measured yet
M1 Pro 32 GB 5 min 50 s480pnot measured yetnot measured yet
M2 Pro 16 GB 4 min 55 s480pnot measured yetnot measured yet
M2 Pro 32 GB 4 min 20 s480pfew Macsnot measured yetnot measured yet
M3 Pro 18 GB 4 min 35 s480pnot measured yetnot measured yet
M3 Pro 36 GB 2 min 40 s480pnot measured yetnot measured yet
M4 Pro 24 GB 3 min 40 s480pnot measured yetnot measured yet
M4 Pro 48 GB 5 min 50 s576pnot measured yet11 minfew Macs
M4 Pro 64 GB 5 min 30 s576pnot measured yetnot measured yet
M5 Pro 24 GB 1 min 35 s480pnot measured yetnot measured yet
M5 Pro 48 GB 2 min 30 s576p3 min 30 sfew Macs2 min 30 s
M5 Pro 64 GB 2 min 30 s576p3 min 50 s3 min 5 sfew Macs
M1 Max to M5 Max
M1 Max 32 GB 3 min 45 s480pnot measured yetnot measured yet
M1 Max 64 GB 5 min 40 s576p10 min 50 s11 min 40 s
M2 Max 32 GB 2 min 45 s480pnot measured yetnot measured yet
M2 Max 64 GB 4 min 20 s576p6 minfew Macs8 min 20 sfew Macs
M2 Max 96 GB 3 min 35 s576pnot measured yet6 min 30 sfew Macs
M3 Max 36 GB 2 min 35 s480pnot measured yet9 min 20 sfew Macs
M3 Max 48 GB 4 min 20 s576pfew Macsnot measured yetnot measured yet
M3 Max 64 GB 4 min576p5 min 50 sfew Macsnot measured yet
M4 Max 36 GB 2 min 15 s480pnot measured yetnot measured yet
M4 Max 48 GB 3 min 5 s576pnot measured yetnot measured yet
M4 Max 64 GB 3 min 10 s576p4 min 20 s4 min 40 s
M4 Max 128 GB 2 min 55 s576p9 min 10 s6 min 30 s
M5 Max 36 GB 55 s480pnot measured yetnot measured yet
M5 Max 48 GB 1 min 20 s576pnot measured yetnot measured yet
M5 Max 64 GB 1 min 20 s576p2 min1 min 40 sfew Macs
M5 Max 128 GB 1 min 20 s576p2 min 35 s2 min
M1 Ultra to M3 Ultra
M1 Ultra 64 GB 3 min 20 s576pnot measured yetnot measured yet
M1 Ultra 128 GB 2 min 45 s576p4 min 35 sfew Macsnot measured yet
M2 Ultra 64 GB 2 min 50 s576p4 min 10 sfew Macsnot measured yet
M2 Ultra 128 GB 2 min 25 s576pfew Macsnot measured yetnot measured yet
M2 Ultra 192 GB 1 min 50 s576pfew Macsnot measured yetnot measured yet
M3 Ultra 96 GB 1 min 50 s576p2 min 30 sfew Macsnot measured yet
M3 Ultra 256 GB 1 min 40 s576p3 minfew Macsnot measured yet
M3 Ultra 512 GB 1 min 50 s576p3 min 40 sfew Macs5 min 20 sfew Macs
M5 Ultra 96 GB 50 s576pfew Macsnot measured yetnot measured yet
M5 Ultra 256 GB 40 s576pnot measured yetnot measured yet

Phosphene also shows an estimate for your own Mac before you press Generate.

Every chip, setting and length
05 / 07

How much disk space do I need?

Memory and disk are separate. Setup downloads no models; you pick them inside the app, and each one downloads once. Downloads resume if the connection drops, and once a model is installed you can work offline. In the Mac app you can keep the models on an external drive formatted as APFS or exFAT.

For a first video: about 18 GB free for setup, then 27.5 GB for the LTX-2.5 models.

Disk space per model
ModelDownload (GB)Free space needed (GB)
App runtime (setup) 0.5 18
LTX-2.5 Q4 video models, with the prompt encoder 27.5
Feature extras (Enhance, Storyboard planner, Upscale & Face Fix, Motion Control)optional up to 10.7
LTX-2.5 High add-on (High, keyframes, Extend)optional 29.5
LTX-2.5 Q8 weights (trained characters)optional 30.02
MiniMax H3 (includes its compact engine build)optional 75 97
H3 Flex adapteroptional 2.8
MiniMax-H3-Image (converted on your Mac)optional 68.6 (keeps about 21) 30
P-H3-Imageoptional 17.5 24
FLUX.2 [klein] 4Boptional 4.3
FLUX 3 Klein (Q4 to Q8)optional 5.3 to 9.9
Ideogram 4optional 28
Music Studio (YuE2)optional 11 (+2.8 for covers) 14
Speechoptional 6 8
LTX-2.3 and the trainer (for training only)optional about 20 + 21
06 / 07

Not supported

Phosphene runs open models with MLX on Apple Silicon. Linux on M1 and M2 is a community-maintained port with best-effort support.

07 / 07

Questions people ask

Can a 16 GB Mac run AI video?

Yes. On a 16 GB Apple Silicon Mac, Phosphene makes LTX-2.5 video with sound from text or a picture at up to 768 px on the long side, plus MiniMax H3 clips at Draft and Standard, FLUX.2 pictures, songs and speech. An LTX-2.5 clip peaks at about 5.7 GB of memory, about 6.7 GB with LoRAs. A 5-second Balanced clip took a median of 8 min 20 s on a 16 GB M4 and 3 min 55 s on a 16 GB M5, model loading included. Close other apps and browser tabs first.

Will it run on my 8 GB Mac?

Phosphene is built for 16 GB and up, so 8 GB Macs are not supported. MiniMax H3 and Music Studio refuse to install below 16 GB, and speech needs 16 GB too. LTX-2.5 is not blocked, and some 8 GB Macs do finish short clips: a 5-second Balanced clip at 480p took a median of 13 min 20 s on 8 GB M1 Macs (8 Macs). Expect long waits and heavy swapping, and quit everything else first.

How much memory does MiniMax H3 (Hailuo) need on a Mac?

16 GB. From 16 GB, MiniMax H3 makes 3 and 5-second clips at Draft and Standard; from 18 GB also High (1024×576). 10 and 15-second clips and Clean up sound need 36 GB. From 60 GB you can also choose H3's full bf16 engine. Below 36 GB, H3 reads the model from your drive as it renders, so the clip comes out the same. A new install needs about 97 GB free: the 75 GB download plus a 22 GB compact engine built on your Mac.

How much memory does LTX-2.5 need on a Mac?

16 GB for text to video and picture to video at up to 768 px on the long side. 48 GB for full size (1280×704), High quality, keyframes, Extend and trained characters. 64 GB for High · 720p, which peaks near 50 GB. From 80 GB keyframes and Extend go up to 1024 px, and from 120 GB there are no size limits.

Does it run on a MacBook Air or a Mac mini?

Yes, if it has 16 GB of memory or more. MacBook Air, Mac mini and iMac use the base chips (M1 to M5), which run everything in their memory size but take longer than Pro and Max chips: a 5-second Balanced clip took a median of 8 min 20 s on a 16 GB M4 and 7 min 50 s on a 24 GB M4. A Mac mini with an M4 Pro took 3 min 40 s with 24 GB.

Can an M1 Mac run it?

Yes, with 16 GB of memory or more. A 5-second Balanced clip took a median of 13 min 20 s on a 16 GB M1, 6 min 10 s on a 16 GB M1 Pro and 3 min 45 s on a 32 GB M1 Max. M1 Macs run the same features as newer Macs with the same memory; they take longer.

How long does a 5-second AI video take on a Mac?

It depends mostly on the chip family. A 5-second Balanced clip with sound took a median of 55 s on a 36 GB M5 Max, 2 min 15 s on a 36 GB M4 Max, 3 min 40 s on a 24 GB M4 Pro and 8 min 20 s on a 16 GB M4, model loading included. The first clip after starting takes longest, because the model loads first.

How much disk space do I need?

About 18 GB free for setup, then 27.5 GB for the LTX-2.5 video models. Everything else is optional and downloads once when you choose it: MiniMax H3 is about 75 GB (97 GB free to install), Music Studio 11 GB, speech 6 GB, the LTX-2.5 High add-on 29.5 GB and its Q8 weights 30 GB. In the Mac app you can keep the models on an external drive: choose it at setup or later in Settings > Models.

Does it run on Intel Macs or Windows?

No. Phosphene is built on MLX for Apple Silicon, so it needs an M1 or newer Mac. Intel Macs and Windows PCs are not supported.

How much memory should my next Mac have?

It depends on what you want to make. 16 GB covers LTX-2.5 at smaller sizes, MiniMax H3 at Draft and Standard, pictures, songs and speech. 24 GB adds character training and picture editing. 36 GB adds 10 and 15-second H3 clips and MiniMax-H3-Image. 48 GB adds full-size LTX-2.5, High, keyframes and Extend. 64 GB adds High · 720p and the full training recipe. For speed, a Pro or Max chip matters more than the year.

How much memory does picture editing need?

Reference Edit (Qwen-Image-Edit) runs from 24 GB at its smallest size, and MiniMax-H3-Image, which edits and combines up to 6 pictures, from 36 GB with MiniMax H3 installed. On a 16 GB Mac, Phosphene makes pictures from text with FLUX.2 [klein] 4B.

Can I train my own character on my Mac?

Yes, an original character or yourself with documented consent: from 24 GB with a lighter recipe (fewer steps, lower rank). The full recipe, measured for character consistency, needs 64 GB. Rendering a trained character needs 48 GB and the Q8 weights. Training uses LTX-2.3, about 20 GB plus about 21 GB for the trainer.

How do I find out how much memory my Mac has?

Open the Apple menu, choose About This Mac and read the Memory line. Phosphene also reads it when it starts and only offers what fits.

Your Mac fits?

Phosphene is free and open source. Download it, pick your models and make your first clip.

Download Phosphene