For more LoRAs and updates, Join my Discord:
https://discord.gg/ZVWVhT43GW
You can deploy MiniMax on Runpod with my template:
👉 https://get.runpod.io/minimax-template
👉 Deployment tutorial
Eight workflows included
Custom prompt - You write the prompt (T2V/I2V)
Auto Prompt - Automatic prompt writing using OpenRouter (T2V/I2V)
Custom Prompt - 2x Upscale (T2V/I2V)
Auto Prompt - 2x Upscale (T2V/I2V)
All workflows and models are preinstalled on my MiniMax Runpod template: https://get.runpod.io/minimax-template
Models:
Diffusion Model:
minimax_h3_fl2va_pruned_int8_convrot.safetensors
Text Encoder:
qwen3vl_32b_h3_ultra_uncensored_heretic_int8_convrot.safetensors
VAE:
minimax_h3_video_vae_fp16.safetensors
minimax_h3_audio_vae_fp32.safetensors
LoRAs:
minimax_h3_fl2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors
Tiny VAE:
taeh3.safetensors
For the Auto Prompt versions, get a key at https://openrouter.ai/keys and paste it into the API Key node. Without it no prompt is written and nothing generates. If a prompt comes back rejected, reword it and run again.
Description
FAQ
Comments (16)
Should I swap the turbo lora to the newer 4step or 8step 1.0? or keep the 0.1 version
8step 1.0 is great
Works great!
Can this be done with R2V?
noob question: where do you add the lora in the minimax workflow?
Not sure what workflow you are using but you can try put the load LoRA as always. In the middle of the model and the KSampler. so would be: Load Diffusion Model → Load LoRA → KSampler.
Im having issues with the resolution. Its like super strechted. My initial pic is 271x979. Maybe thats the problem?
Try connecting your input resolution to the video resolution instead of using a resolution selector
Really hope people get this working for them so we can drown out the fat fetish freak in the gallery, lol
idk man that fat physics is pretty fucking impressive,lol
where do you specify the length of the video?
There's an integer input I believe the default is 6.0
The last two times I've tried to download the Text Encoader, it fails at the very end. I'm left with a 24.5 GB "Unconfirmed 183702.crdownload". Whats happening here? I've never had a download fail at the very end.
I had the same problem before. Use the Free Download Manager browser extension and the app itself. It worked for me. If you get an error while downloading, just restart the download and it will resume where it started.
Find the incomplete file that ends with .crdownload and rename it to the exact final filename of the model (the full .safetensors name).
Open PowerShell and navigate to the folder where that file is located.
Run the hf download command for the model and filename, pointing it to the current folder. "hf download ethanfel/Qwen3-VL-32B-Ultra-Heretic-H3-ComfyUI-INT8-ConvRot qwen3vl_32b_h3_ultra_uncensored_heretic_int8_convrot.safetensors --local-dir ."
Wait until it confirms the download is complete.
This usually happens with Chrome for large files. I use Edge and it downloads without any problems.
some good recomendations, I just DL with hf cli to avoid the problem on 20+ GB