SenseNova-U1.5 ConvRot for ComfyUI: 50GB Model on a 12GB GPU

ComfyUI Wikinews

ConvRot quantization runs SenseNova-U1.5-8B-MoT in ComfyUI on a 12 GB GPU: INT8 (17.6 GB) and hybrid W4A8 (13.8 GB) checkpoints plus an 8-step speed LoRA.

SenseNova-U1.5-8B-MoT ConvRot quantization (weights | custom node) is a community release that brings the 50 GB bf16 any-to-any model (text-to-image, image editing, multi-reference) down to consumer GPUs in ComfyUI. Two ConvRot checkpoints, an INT8 build and a hybrid W4A8 build, run on an RTX 4070 12 GB at 2048×2048, and the official 8-step speed LoRA is included.
Same-seed A/B: bf16 reference, INT8 ConvRot and hybrid W4A8 results

What is in the release

Two quantized checkpoints plus the official speed LoRA:

FileSizeFormatNotes
SenseNova-U1.5-8B-MoT-T8-int8-convrot-tagged.safetensors17.58 GiBINT8 ConvRotRecommended, maximum fidelity
SenseNova-U1.5-8B-MoT-T8-hybw4a8-L18-41.safetensors13.80 GiBHybrid INT8 + W4A8Layers 0-17 in INT8, layers 18-41 in true W4A8
SenseNova-U1.5-8B-MoT-LoRA-8step-ComfyUI.safetensors0.76 GiB8-step speed LoRAOfficial LoRA converted for ComfyUI

Both checkpoints go into ComfyUI/models/diffusion_models/SenseNovaU1.5/, and the LoRA into ComfyUI/models/loras/. Inference runs through a ConvRot-aware fork of the T8 wrapper node, which adds explicit activation rotation and manual dequantization for INT8, kernel routing for W4A8 through Comfy-Org's comfy-kitchen, and runtime guards that stop ComfyUI's weight streaming from corrupting packed quantized tensors.

Quality

Measured against the bf16 original with same-seed full-pipeline A/B runs:

bf16 (reference)INT8 ConvRot (same seed)Hybrid W4A8 (same seed)
bf16 referenceINT8 ConvRotHybrid W4A8
Reference0.43% pixel diffVisually indistinguishable

The INT8 variant reports a 0.43% pixel difference over a full-pipeline same-seed A/B, with per-layer weight reconstruction error well under 2%. The hybrid reports about 7% relative L2 error on its 4-bit layers (Lloyd-Max codebook, group size 16, FP8 group scales) yet is visually indistinguishable from bf16 in same-seed tests. Same-seed trajectories are chaotic, so the authors present the images as quality samples rather than pixel comparisons.

Why the hybrid

The interesting finding behind the release: SenseNova-U1.5 tolerates true W4A8 activation quantization in its later layers but not in its earliest ones. Quantizing the first transformer blocks destroys prompt coherence, while layers 18+ quantize to W4A8 transparently. The authors located the boundary empirically with a bisect ladder of hybrid checkpoints, then produced the hybrid files by byte-level merging of two independently validated checkpoints, so no layer was ever re-quantized during the merge.

Performance

On an RTX 4070 12 GB, both variants run fast even though the weights exceed VRAM: ComfyUI streams weights on demand, and the quantized formats move 3-4x fewer bytes per step while computing through fast integer tensor-core kernels, so the overflow never becomes a slowdown. bf16 on the same card streams about 47 GB per step and is drastically slower.

Getting started

  1. Install the custom node: git clone https://github.com/Milor123/ComfyUI-SenseNova-U1.5-ConvRot into ComfyUI/custom_nodes/, with comfy-kitchen >= 0.2.31 installed (tested with ComfyUI commit 82f839f5).
  2. Download the recommended INT8 checkpoint (or the hybrid) into ComfyUI/models/diffusion_models/SenseNovaU1.5/ and the speed LoRA into ComfyUI/models/loras/.
  3. Run one of the example workflows from the repo's examples/ folder:

Comments

Sign in with GitHub to join the discussion.

Loading comments…
SenseNova-U1.5 ConvRot for ComfyUI: 50GB Model on a 12GB GPU | ComfyUI Wiki