Downloads · 30 days
12.4K
21% of all-time downloads
drbaph/vdn-minimax-h3-int8-convrot-comfyui
vdn-minimax-h3-int8-convrot-comfyui is a machine learning model from drbaph. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for minimax-h3. The card lists the license as other.
Pre-quantized Comfy Kitchen INT8 ConvRot version of the 8-step checkpoint (stage-dmd-step-250) from OpenVDN/vdn-minimax-h3. For use with ComfyUI-VDN-H3.
Downloads · 30 days
12.4K
21% of all-time downloads
All-time downloads
59.3K
Public
Repo size
3.5 GB
Likes
29
Public
Click a slice to open those files.
.safetensors3.5 GB · 100%
From the Hugging Face model README
stage-dmd-step-250 — INT8 ConvRot (ComfyUI pre-quantized)Pre-quantized Comfy Kitchen INT8 ConvRot version of the 8-step checkpoint
(stage-dmd-step-250) from OpenVDN/vdn-minimax-h3.
For use with ComfyUI-VDN-H3.
<video src="https://huggingface.co/drbaph/vdn-minimax-h3-int8-convrot-comfyui/resolve/main/assets/int8-ref2va-8step-vdn-1.mp4" controls loop width="100%"> </video>
Requires ComfyUI-VDN-H3 v1.3.0+ (release).
Same weights, same adapters, A/B-verified visually identical output. What you save:
| original stage | this repo | |
|---|---|---|
| linear branch | 4.28 GB bf16 | 2.20 GB int8 |
| stage total | 5.09 GB | 3.25 GB |
branch VRAM (cache_gpu) | ~4.3 GB | ~2.2 GB |
| peak VRAM while loading | — | ~4.7 GB lower |
Faster: the branch's dispatched matmuls run 2.7x faster than bf16, and the A/B render finished ~1.2x faster end-to-end (~111 s -> ~95 s, single run each).
8 steps, er_sde / beta, 1280x736 / 61 frames, merge, cache_gpu, same seed.
Identical output; wall clock from a single run each (±5 s sampling).
| stage | precision | wall time | peak VRAM (min free) | video |
|---|---|---|---|---|
stage-dmd-step-250 | bf16 | ~111 s | 3.6 GB free | <video src="https://huggingface.co/drbaph/vdn-minimax-h3-int8-convrot-comfyui/resolve/main/assets/ab_bf16_8step_er_sde_beta.mp4" controls loop width="100%"> </video> |
stage-dmd-step-250-int8_convrot_comfyui | int8 convrot | ~95 s | 8.3 GB free | <video src="https://huggingface.co/drbaph/vdn-minimax-h3-int8-convrot-comfyui/resolve/main/assets/ab_int8convrot_8step_er_sde_beta.mp4" controls loop width="100%"> </video> |
Download this repo into ComfyUI/models/vdn/ (the folder name becomes the vdn_checkpoint entry), then select it in the Apply-VDN node's
vdn_checkpoint dropdown. Nothing else changes; the base MiniMax-H3 model and
this node's settings work exactly as before.
Rendered with this repo's INT8 ConvRot stage on a ref2va base: 8 steps,
er_sde / beta, 928x928.
<video src="https://huggingface.co/drbaph/vdn-minimax-h3-int8-convrot-comfyui/resolve/main/assets/vdn_h3_smoke_00046_.mp4" controls loop width="100%"> </video>
tools/quantize_vdn_branch_int8.py from the ComfyUI-VDN-H3 repo, using
comfy-kitchen's TensorWiseINT8Layout (convrot, per-channel) — the same
serialization ComfyUI ships for its own int8-convrot diffusion models. Only the
branch's dispatched linear weights are quantized; adapters, biases, norms and
convs pass through untouched.
VDN-H3 is built on MiniMax-H3 and starts from its released transformer weights. We also thank Diffusers, FlashAttention, and Triton, which the optimized inference path is built with.
VDN-H3 is a derivative of MiniMax-H3 and is distributed under the MiniMax H3 Community License Agreement, included here verbatim from the upstream repository.
The agreement grants rights only in its applicable territory, defined as worldwide excluding the European Union, the United Kingdom, the Republic of Korea, and the United States of America. It states that use outside the applicable territory is not authorized and invites people in an excluded territory to contact MiniMax about obtaining a license.
The agreement also contains redistribution requirements and an Acceptable Use Policy. Among other requirements, a distribution must include the agreement, modified files must carry notices of their modification, and distributions to third parties other than through hosted services must include the NOTICE file supplied with the code repository. Please read the agreement in full before using or distributing VDN-H3. This note is not a substitute for the license text or for legal advice.