saman

i am anesthetist, tech + film + game ♥️

[AI Search @aisearchio]

Acceleration loras for Minimax H3 ref2va

https://t.co/sOryWiaJ7q

[ノマ扉 | nomadoor @noma_door]

構造化プロンプト元祖(多分)の後継、Fibo 2.0 の workflow です
Dev と Edit Turbo、画像生成・画像編集・プロンプト変換、全部置いてあります
https://t.co/ac6S7TOhWO

構造化プロンプトが売りといいつつ、適当な自然文でもそこそこ効くのはよいですね。ね、FLUX 3さん(´・ω・｀)

[有趣的80后程序员 @sunbaolong_2001]

Swap ONLY ONE person in a duet dance with an anime character! 🎬✨
H3 Character Swap LoRA v1.5 (for MiniMax H3) allows for surgical character replacement in ComfyUI:
✅ Real + Anime duet in ONE continuous shot
✅ 100% motion, camera path, and lighting sync
✅ Expression transfer (micro-gazes, eyebrow raises, smiles)
✅ Paired with SAM 3.1 for pinpoint crowd & multi-person isolation
Full guide & RunningHub workflows below: 👇
🔗 https://t.co/TNDQIHT9Qg
#MiniMaxH3 #ComfyUI #CharacterSwap #AIVideo #SAM31 #GenerativeAI

[Kamimoto(かみもと) @sep_is_heim]

【MiniMax-H3】LightX2Vが10/20にTurbo LoRAを更新！

FL2VA：4step v1.3／8step v1.1
Ref2VA：4step v1.0／8step v1.1

768p向けの推奨Shiftは映像6、音声3。従来の544p向け設定とは異なるので注意。何が変わったかは分からないけどね！

https://t.co/fuOgM6mXYL

[NRDX @Alissonerdx]

Sorry folks, I know you're tired of Character Swap LoRAs 😅 but here's an update to my v1.0.

I spent the last week training version after version to improve likeness, and I think I landed on a really nice one. I almost called it v2.0, but for that I'd want multiple characters swapped at once. I don't doubt it's possible, I just haven't tested it. So v1.1 it is.

It's also much smaller than the previous one (80 MB vs 620 MB): I extracted the result with SVD at a dynamic rank, keeping 95% of the energy (Frobenius norm) per layer. So every layer has its own rank: the lowest is 4 and the highest is 34 (median 7).

Hope it really beats v1.0. Now back to my OpenJutsu for the Comfy challenge 🔥

Links:

https://t.co/YuJEujqJxt

https://t.co/wM35b95Avu

If this LoRA proves useful for your work in any way, please consider helping me continue training more LoRAs like this one ☕ https://t.co/5Wv2ZzzWVE

Below are some tests I ran at a low resolution (0.3 MP); as a result, you might notice some artifacts and the like, which is perfectly fine, and likely due to the very low resolution.

#ComfyUI #Minimax #H3 #aigenerated #CharacterSwap @ComfyUI @MiniMax_AI @huggingface

[AI Search @aisearchio]

MOSS Image 8 step finetune is here. ~5x speedup compared with the base model (50 steps).

https://t.co/z8LrZ4vZUX

[CadX Studio @CadX_Studio]

Your AI can design.
Can it design for manufacturing?

CadXStudio 2.0 generates DFM-ready models. One prompt → an injection-moldable gaming mouse:

• 7 parts, 1.8 mm walls
• Snap-fits + alignment pins
• 0.5° rib draft

All in your browser: https://t.co/gkW2OZcbdd
Prompt + STEP 👇

[Forha @Forhanvv]

From 𝟭𝟭 𝗺𝗶𝗻𝘂𝘁𝗲𝘀 down to 𝟳𝟮 𝘀𝗲𝗰𝗼𝗻𝗱𝘀.

A new ComfyUI node just cracked real-time speed for 𝗠𝗶𝗻𝗶𝗠𝗮𝘅 𝗛𝟯 character video generation.

𝗖𝗼𝗺𝗳𝘆𝗨𝗜-𝗥𝗲𝗳𝟮𝗩𝗔-𝗩𝗦𝗔 achieves a massive 𝟵𝘅 𝘀𝗽𝗲𝗲𝗱𝘂𝗽 using sparse attention DiT:

• 𝟳𝟮𝘀 𝘄𝗮𝗹𝗹 𝘁𝗶𝗺𝗲: Full 5.16s video @ 24fps (124 frames, 1344×768)
• 𝟭𝟯.𝟱 𝗚𝗕 𝗣𝗲𝗮𝗸 𝗩𝗥𝗔𝗠: Down from ~22.5 GB (runs on 16GB GPUs)
• 𝟰-𝗦𝘁𝗲𝗽 𝗦𝗮𝗺𝗽𝗹𝗶𝗻𝗴: Euler/simple sampler with 75% video sparsity
• 𝗭𝗲𝗿𝗼 𝗙𝗮𝗰𝗲 𝗗𝗿𝗶𝗳𝘁: Reference tokens stay locked out of attention pruning to maintain 100% character identity
Benchmark comparison on RTX 4090: → Native Dense H3: ~650s (~11 min)
→ Video Delta Net (VDN): ~135s
→ Ref2VA VSA: ~72s

Open-source on GitHub:
https://t.co/K5vcI6CeVh

[AI Search @aisearchio]

Now THIS is a good image model. MOSS Image, 5B Apache 2.0.

Already very uncensored!

https://t.co/vmb00jZF8q
https://t.co/bZIDDUkBUx

[Enze Xie @xieenze_jr]

🎬 MiniMax-H3, now powered by Sol Engine in ComfyUI!

Generate 5s videos with native audio on a single GPU:
⚡ RTX 5090: 35.5s
⚡ DGX Spark: 64.3s

Meet Sol-H3 Studio: a visual workspace for local video generation, powered by our optimized inference engine.

Open source 👇
https://t.co/OYZxIPgxwU

[Brennen Herbruck @bherbruckx]

SolidCraft v0.14.0 is out 🚀

https://t.co/77tel1ocdo

- Import FreeCAD, Rhino & glTF
- Gears, rack & pinion, IK, Motion Trace
- Sketch patterns, infinite lines, Split Body
- AI chat: queue, stop, model picker, Grok
- In-app update alerts
- Tons of fillet/boolean fixes

[Brennen Herbruck @bherbruckx]

AI builds a waterproof Raspberry Pi 5 enclosure in SolidCaft

https://t.co/77tel1oK2W

[Brennen Herbruck @bherbruckx]

Planetary gearbox, built by an agent in SolidCraft. 12-tooth sun, three planets, a 48-tooth ring, real joints, and a motion study at the actual 5:1 ratio.

https://t.co/77tel1ocdo

[Brennen Herbruck @bherbruckx]

Free and open-sourced parametric CAD SolidCraft works with AI to create a workbench with Ikea countertop and 4040 aluminum extrusions

Try it out:
https://t.co/77tel1ocdo

[akatz @akatz_ai]

H3 Character Swap Lora V1.5 is out!

Better identity preservation + better handling of cuts + a few improvements over V1

AND I'm releasing the workflow/tool I've been using with it too👇

It's not V2 yet 😅 I still need more time to cook. Several revisions haven't passed my hiiigh V2 standards. But V1.5 did beat out V1 in my official "trust me bro" benchmark below.

Lora: https://t.co/1JDtSLHdTt

[atomic.chat @atomic_chat_hq]

Run Ternary Bonsai 2 locally in Atomic Chat 🪴

@PrismML shrank Qwen3.5 7x so the 9B goes from 17.9 GB to just 2.42 GB and it keeps up or outperforms the original model on benchmarks

Run AI models locally -> https://t.co/RbcCOIht9R

[ITOK @ITOKtw]

MiniMax H3，lightx2vのlight vaeにComfyUIが対応しました。デコード時間半減！
int8:13.28s
lynnreal：10.14s
lightvae-lite-h3 int8オリジナル：5.71s
lightvae-lite-h3 int8 Kijai版：5.85
オリジナル：https://t.co/gHiGIULTeO
Kijai版：https://t.co/CInW4M3asZ 1.04GB

[PC Master Race Latinoamérica @PCMR_LATAM]

Shadow of the Colossus PS4 Remake now has an unofficial native PC port! 🎮

60 FPS, 4K output, FSR, keyboard & mouse, Xbox/PS controllers, and Vulkan 1.3 support. 🗿

https://t.co/FLHAUm43ON

[0xMarioNawfal - The Roundtable Show @RoundtableSpace]

CapCut now has a free, open-source replacement. No watermark, no account, no subscription.

Concat is a video editor for Windows, macOS, Linux and Android that covers what most people open CapCut for:
→ Auto-captions, generated on your own computer
→ Text-to-speech, plus voice cloning from a few seconds of audio
→ Background removal for people and objects
→ Keyframe animation with a curve editor
→ 170+ effects, filters, transitions and text animations
→ Multi-track editing and 4K export

Everything runs locally. The AI models for captions, voices and cutouts download once and then work offline. Your footage never leaves your computer.

It's in beta, so expect a few rough edges.

https://t.co/olfuPoupJt

[有趣的80后程序员 @sunbaolong_2001]

Skip 76.8% of attention compute & save 33 SECONDS per video! 🚀⚡
Veda Attention is officially the fastest sparse attention mechanism for MiniMax H3 in ComfyUI:
✅ Uses a dedicated neural predictor to calculate only 23.2% of attention blocks
✅ 10s Singularity render: 178s ➔ 145s (19% time saved on RTX 4090!)
✅ Works seamlessly with FastH3 V2 & Singularity
⚠️ CRITICAL TIP: Incompatible with FastH3 Trim (42 layers vs 50 layers)!
Full breakdown & RunningHub workflows below: 👇
🔗https://t.co/zAjWnKJ3BQ
#VedaAttention #MiniMaxH3 #ComfyUI #AIVideo