Minimax H3 🤺⚔️ 🥊Fight Flow Fight Interaction LoRA Better attack-response causality, more natural initiative changes, stronger state continuity, and fewer “attack → reset → attack again” sequences.

Qwen Image2.1 🧑‍🌾Correcting Background Blending LoRA V2 keep the subject position and size unchanged, adjust the background perspective angle to align with the subject's vanishing point. seamless blending between subject and background.

🚀 Meet Qwen-Image-2.1-Turbo — create and edit images in just 8 denoising steps! Open weights now available! Built on Qwen-Image-2.1, Turbo is an accelerated checkpoint on the same 7B visual generation architecture. Fewer steps does not mean lower quality: it still generates strong 2K images from text, and supports continued creation through natural-language edits, from adding accessories to changing a scene. Start directly with Diffusers: load QwenImage21Pipeline and the checkpoint’s recommended 8-step sampling schedule is ready to go. 🌐 Prefer an API? Qwen-Image-2.1 Pro and Turbo APIs are now officially live. Choose a hosted API for your application, or use the Turbo weights in your own workflow. 🤖 ModelScope: 🤗 Hugging Face: 🔗 Pro API: 🔗 Turbo API: What will you create with 8 steps? Share your results with us!

Qwen-Image-2.1 ➡️🖼️Next-Scene-LoRA for cinematic scene continuation: it takes an input image and a director-style instruction and generates the next scene camera moves, reveals, lighting shifts while keeping the world, characters and style consistent

New Release: Wasserman's 80s Horror Lora My free LoRA for MiniMax H3. One trigger word and your scene looks like a 1984 rental: VHS or grindhouse 16mm for 70s style. One-line ComfyUI install. Be kind, rewind. 🤡🔪📼 Repo here👇 What you get: 🎞️ VHS or 16mm from one trigger word 🧛 Your scene stays your scene. It changes the look, not the story 🔊 Trained with audio, so the screams, dialogue and creaky floorboards still hit 🧩 Plug-and-play ComfyUI workflows: text-to-video and image-to-video ⚡ A one-line installer that finds ComfyUI, drops in the LoRA and loads the workflows 🩸 Works with standard or uncensored H3 (uncensored hits hardest) I trained it on 48 horror shots I made over the course of a few weeks. If you saw my horror montages, this is trained on those.

Qwen Image2.1 🔂📸 Multi-Angles camera-control LoRA (V1/v2) Reshoot any object from a controlled camera position: feed a photo, ask for a different azimuth/elevation, get the same subject from that angle.

Qwen Image 2.1 😃Consistent Lighting lora consistent lighting to the product or object, enhance specular highlights，front-back spatial relationship，add natural contact shadow, and blend it seamlessly into the background

I tried the X2-Detail-VAE for MiniMax H3 on my RTX5090 locally in ComfyUI. One mode's a free upscale at no extra render time, the other's hit-or-miss. - 2X decode: 736×1280 → 1472×2560. Clean upscale, slightly crisper edges, no new detail. - Detail mode: 7–14% darker. A touch crisper in this clip (lantern text, pattern edges), but no consistent gain across my tests, some came out softer. - Render time: ~320s for both. Links below👇

MiniMax H3 🧐 Person Remover LoRA tracks the person with SAM 3.1, fills the mask green, and generates the replacement background in overlapping windows. Supply the original video and a clean version of its first frame.

Meet Iris-3B: a pixel-space T2I model and general vision learner that generates every pixel directly. No VAE. Pre-trained from scratch, 3B, open weights. We also explore its generative prior on detail-critical vision tasks: depth estimation and image restoration. Fully open: weights, code and demo (Apache 2.0)

😌 Minimax_h3 4step_full_critic and Minimax_h3 4step_lora_critic for ComfyUI These DMAD 4-step models compress that pipeline down to just 4 steps without causing the generation quality to fall apart

Minimax H3 🙂 SOL fl2va pruned By comfyanonymous -forces the model's Query/Key/Value (Q/K/V) tensors to pipe through the optimized Sol-Attn backend -cuts down generation time and lets you render video/audio blocks much faster per second of video gen 👇

Tencent 🤨 📹dropped Prism -Opensource Video+Audio Gen Model -Dynamic Sparse Attention for 2K Video-Audio Generation -Supports native joint video-audio generation at 720p, 1080p, and 2K resolutions -Native joint video-audio training 👇

Qwen-Image-2.1 📦 Voxel-Style LORA designed to transform input images into a distinctive voxel-art style with block-based geometry, simplified forms, low-resolution 3D structures, and a stylized digital aesthetic.