ahegao.redahegao.redComfyUI workflowsahegao.red home

ComfyUI · NSFW · 18+

MiniMax-H3 NSFW video workflow

The Video Pack runs on MiniMax-H3, the open image-to-video model by MiniMax, packaged for ComfyUI through Comfy-Org on Hugging Face. The pack ships the pruned int8 reference-to-video build, the Qwen3-VL 32B text encoder, separate video and audio VAEs, and the ref2v turbo LoRA. A source image and a prompt produce a clip of 5 to 15 seconds whose soundtrack matches the motion.

What the pack ships

ComponentFileNotes
Diffusion modelMinimax-h3 ref2va, pruned, int821 GB, needs a 32 GB card
Text encoderQwen3-VL 32B (nvfp4 awq)Runs on Ampere up, no Blackwell required
Video VAEminimax_h3_video_vae, int8Encodes the frames
Audio VAEminimax_h3_audio_vae, fp32Generates the soundtrack
Turbo LoRAminimax_h3_ref2v_turbo_4stepShips applied, speeds sampling
Tuned LoRA1 quality LoRA from our CivitaiLocks in the final look

Everything installs with one command. Sampler, CFG, and LoRA settings arrive pre-calibrated. Leave them as they ship.

How a clip gets made

  1. The source image becomes the first frame of the clip.
  2. The diffusion model animates it, 24 fps, 5 to 15 seconds.
  3. The audio VAE generates the soundtrack in the same pass, so sound matches motion.
  4. A two-stage quality pipeline renders, then refines the result.

The prompt carries two blocks: one for the picture, one for the sound. Describe one continuous shot in temporal order: what moves, then the lighting and camera, then the matching sounds.

Writing prompts that hold up

  • Describe motion, not appearance. The source image already sets the look.
  • One primary action per clip. Layered actions lower coherence.
  • Do not contradict the source image. The clip starts from it.
  • One continuous action scales better than a sequence of position changes.

Hardware

NVIDIA GPU with 32 GB+ VRAM (RTX 5090). On RunPod, RunPod RTX 5090 / RTX 5000 Ada / A6000+. Disk needs 60 GB for models. Setup guides are linked below.

Adults only (18+). All sample output is synthetic, computer-generated fiction. Last verified October 2026.

Keep reading

PackImage PackGenerate photorealistic uncensored scenes and edit existing photos in one workflow, with identity-preserving editing. $49 one-time.PackVideo PackTurn any image into a clip with sound matched to the motion; motion-synced audio and likeness-preserving two-stage pipeline. $69 one-time.PackImage + Video PackBoth pipelines in one bundle: renders a high-resolution image, then animates it into a clip with sound while character and setting stay identical. $89 one-time.GuideComfyUI NSFW local setupInstall an ahegao.red ComfyUI workflow pack on Windows or Linux: one command installs Docker ComfyUI and every model file. Generate uncensored images and video on your own GPU.GuideComfyUI NSFW on RunPodRun an ahegao.red ComfyUI workflow pack on a rented RunPod GPU. Create a network volume, add credentials, deploy the official template, and connect on port 8188.GuideBest GPUs for NSFW AI generationWhat GPU do you need for uncensored AI image and video generation? VRAM requirements per pack, a card ladder, and when renting a RunPod card beats buying.Model notesKrea-2 NSFW ComfyUI workflowThe Image Pack runs Krea-2 with an abliterated Qwen3-VL text encoder. Uncensored text-to-image and identity-preserving photo edits on your own GPU.Model notesAbliterated Qwen3-VL: uncensored prompt readingThe ahegao.red packs read prompts through abliterated Qwen3-VL models: the Image Pack ships Huihui-Qwen3-VL-4B, the Video Pack ships Qwen3-VL 32B. Safety filtering removed, fully local.
ahegao.red

© 2026 ahegao.red. All rights reserved.

All images and videos advertised on this page are computer-generated works of fiction. They do not depict any actual person, living or dead. All persons depicted therein are, and are represented as being, eighteen (18) years of age or older.