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
Four workflows: T2V and I2V, each in a Custom Prompt version (you write the prompt) and an Auto Prompt version (you type one line and OpenRouter writes the full H3 prompt for you).
All four come preinstalled with every model above on my 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
Adapting workflow to newest Turbo LoRAs
FAQ
Comments (9)
API key??
Do you have a offline version of it? local?
What is the difference between version 2.0 and version 1.0? The connection status with the node and both seem identical
Just using the new turbo LoRAs
It appears the intended AutoPrompter model requires tokens now? The available free models are censored/moderated from what I can see. Does anyone have a workaround for the AutoPrompt?
Use the non AutoPrompt version.
Gemini flash, Opus, Sonnet, Deepseek, Qwen, Muse Spark are uncensored.
Thanks for the update! Works great!
Your workflow works really well for me. I tested the previous version and it just works flawless, especially the local host prompts.
Question: can your workflow be integrated to do long video? I tried to just a ref video file but it did the full video and just added whatever else I promoted so let's say I added the 5 second reference video and promoted about 5 seconds and it created a 10 seconds video but not a direct continuation. Because the added 5 seconds was part of the new video, I believe it could work but just don't want to add the full reference video though. I hope it makes sense what I am talking. :D
Before this updated workflow, output quality was outstanding, now every video generated looks so weird, full of artefacts, blurry.. How to fix this ?
It has to do with the turbo LoRa I'm pretty sure, but how to fix it ?