Nils Fleischmann (@nifleisch) header
Nils Fleischmann (@nifleisch) profile picture

Building Fast Models @PrunaAI

Munich
Joined November 2023
Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)Photo by Nils Fleischmann (@nifleisch)
Spent the last weeks doing model brain surgery on the MiniMax-H3 video model, turning it into a world-class image model. Everything is open, and we wrote a technical blog post about how we did it. Go read it. pruna.ai/blog/oss-p-h3-image…
2
6
1
14
675
Over 600k downloads in 5 days. Still trying to make sense of that number. Thanks to everyone who tried P-H3-Image 🙏
3
130
if bfl gets scorsese, can we have george lucas as a strategic advisor at pruna? p-h3-image is giving me strong star wars vibes
1
1
6
195
P-H3-Image is surprisingly good at photo grids. The image below is a single 16 MP generation.
1
11
366
like many people, i've wondered my whole life if I could beat a horse sized duck in a Mortal Kombat fight ... you can find out too with P-Video-2-Pro-Reference
Introducing P-Video-2-Pro-Reference based on @MiniMax_AI H3! 🎬 Combine images, videos, and audio references to guide how your video looks, moves, and sounds. Keep characters consistent, transfer motion, and bring different visual elements together in a single generation. • 𝗜𝗻𝗽𝘂𝘁: Text prompt with image and/or video references, plus optional audio • 𝗥𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲𝘀: Up to 9 images, 3 videos, and 3 audio files • 𝗗𝘂𝗿𝗮𝘁𝗶𝗼𝗻: 5–15s at 480p or 768p • 𝗣𝗿𝗶𝗰𝗶𝗻𝗴: Output starting at $0.015/s in Cost mode, with reference inputs billed separately • 𝗖𝗼𝗻𝘁𝗿𝗼𝗹: First and last frames, seed, aspect ratio, and prompt upsampling Two modes: Speed and Cost. Choose the trade-off that fits your workflow. Available via our inference partners @Cloudflare, @scenario_gg, @replicate, @WiroAI, @TellersAI, @eachlabs, @wavespeed_ai, @inference_sh, @_segmind, @togethercompute, @EdenAIco, and @runware Validated by our benchmark partners @datapointai, @RapidataAI, and @DesignArena 📚 Model docs: docs.api.pruna.ai/guides/mod… 🧩 API: dashboard.pruna.ai/login
1
4
9
209
it’s weirdly hard to find an image model whose outputs could pass as a photo off your iphone ... rl for image gen was a mistake 👀
1
3
138
Nils Fleischmann retweeted
P-Video-2-Pro dominates the quality-speed frontier on @DesignArena for both Text-to-Video and Image-to-Video! Image-to-Video: • Quality: #3 · 1325 Elo · 8.0s • Speed: #2 · 1325 Elo · 4.5s Text-to-Video: • Quality: #4 · 1308 Elo · 8.6s • Speed: #6 · 1298 Elo · 4.7s That’s a 19–22 place jump from P-Video-2.
12
23
6
125
10,037
Nils Fleischmann retweeted
Last week we @PrunaAI announced our new model P-Video-2-Pro based on @MiniMax_AI H3. Today we are ranked #2 in Image-to-video and #4 in Text-to-video on @DesignArena. This now shows: good quality does not have to be slow or expensive. Super proud of the team!
3
8
110
This is peak video generation and it comes from Europe! 🇪🇺
P-Video-2 Pro (Quality) and P-Video-2 Pro (Speed) by @PrunaAI tie each other for #2 on Design Arena's Image to Video leaderboard with an Elo of 1325. Based on MiniMax H3 by @MiniMax, P-Video-2 Pro (Quality) generates in 8.0 seconds, while P-Video-2 Pro (Speed) has an 4.5-second generation time, placing it on the Pareto frontier of Preference vs Speed. Congratulations to the entire @PrunaAI team!
3
10
267
Pruna-Qwen-Image-2.1 might be the world's first 5-step distillation LoRA lol.
1
7
245
With well-written prompts Pruna-Qwen-Image-2.1 gets surprisingly close to base. A few side-by-sides below 👇
2
1
1
7
317
Shoutouts to our distillation wizard @simlang_. It was a lot of fun to build together, and I'm proud of how far we got in the limited time we had.
1
3
55