MiniMax-H3-Qwen3-VL-Abliterated-NVFP4

An abliterated, NVFP4 (4-bit) Qwen3-VL-32B text encoder for MiniMax-H3, made with convert-to-quant. It is a drop-in replacement for the H3 conditioning text encoder (ComfyUI CLIPLoader, type minimax). The H3 video weights are unchanged — only the prompt-understanding encoder is modified.

Abliteration

The refusal direction (Arditi et al. 2024) is orthogonalized out of the language layers' o_proj / down_proj and the token embedding (101 tensors edited), which reduces prompt sanitization/refusal so the encoder conditions more faithfully on the input prompt. The video model is unaffected.

Quantization

• Quantized with ctq --nvfp4 --qwen_vlm --comfy_quant (weight-only PTQ, comfy_quant layout).
• Protected (kept bf16): token embeddings, first & last language layers, and the full vision tower (--qwen_vlm filter). The middle language-layer linears go to NVFP4.
• Size: ~17.9 GiB (bf16 source is 48 GiB) — fits a 24 GB consumer GPU with ComfyUI offload.

Quantization fidelity (4-bit vs bf16)

Everything is held constant except encoder precision — the bf16 abliterated encoder vs this NVFP4 abliterated encoder, same prompts, same seeds, same NVFP4 DiT — over a 16-clip paired set.

Verdict — perceptually and temporally equivalent to bf16:

Metric	Value	What it measures
LPIPS ↓	0.099	Learned perceptual per-frame distance
SSIM ↑	0.891	Structural per-frame similarity
Temporal flicker (mean inter-frame diff)	13.75 → 14.12 (+2.7%)	Frame-to-frame stability across the clip set

LPIPS ≈ 0.10 with SSIM ≈ 0.89 means the 4-bit encoder's frames are perceptually the same as bf16's, and the near-identical flicker (13.75 vs 14.12) shows temporal behavior is preserved — the 4-bit conditioning adds no visible shimmer or instability. These are the fidelity conclusion.

FVD — internal relative reference only, not a fidelity verdict. i3d FVD over the same 16 paired clips is 1059.5. This is not a published-scale FVD number: FVD estimates a high-dimensional feature covariance and needs large samples (papers typically use 2048+); at n = 16 the covariance is under-determined, which inflates the absolute value and makes it non-comparable to reported FVD. A meaningful reading would require a same-precision noise floor (bf16 vs bf16 at a different seed); we report the raw figure only for transparency and rely on the perceptual and temporal metrics above.

PSNR over the paired clip is 24.85 dB — low only because a tiny conditioning change makes the diffusion sampler diverge pixel-wise while staying perceptually equivalent, which is exactly why LPIPS/SSIM (not PSNR) are the right lens here.

Verified

Item	Result
Loads in ComfyUI (CLIPLoader type minimax)	Yes
Generation quality after abliteration + 4-bit	Clean 720p; perceptually equivalent to bf16 (SSIM 0.891, LPIPS 0.099)
Character identity-lock (ref2va, 3-view)	Consistent
Peak VRAM, full 4-bit stack, 24 GB budget	~25.7 GiB (runs on RTX 4090 with offload)
Speed	~29 s/step at 24 GB; ~2 min/clip with the 4-step turbo LoRA; faster on 32 GB+

NVFP4 runs fastest on Blackwell; on Ada (RTX 4090) a cu130 torch build enables the comfy_kitchen FP4 kernels (otherwise emulated and ~2.3× slower).

Usage (ComfyUI)

1. Put qwen3vl_32b_minimax_h3_abliterated_nvfp4.safetensors in ComfyUI/models/text_encoders/.
2. Download the companion 4-bit weights:
   • DiT: minimax_h3_ref2va_pruned_nvfp4.safetensors → diffusion_models/ (lilcheaty/MiniMax-H3-NVFP4)
   • VAE: minimax_h3_video_vae_fp16 + minimax_h3_audio_vae_fp32 → vae/ (Comfy-Org/MiniMax-H3)
   • Turbo (optional, 4-step): minimax_h3_fl2v_turbo_4step_v0.1.safetensors → loras/ (Gluttony10/MiniMax-H3-INT8-CONVROT)
3. In the H3 workflow set CLIPLoader to this encoder and UNETLoader to the NVFP4 DiT. An example ref2va identity-lock workflow (h3_ref2va_multiview_facelock.json) is included.

Responsible use

Not-for-all-audiences research artifact for fictional characters only. Abliteration lowers general refusal; it does not permit — and you must not generate — sexual content involving minors (including fictional/AI), non-consensual imagery, real identifiable people without consent, or any illegal content. Mark synthetic outputs and follow the MiniMax Community License and local law.

Credits & license

• Base video model: MiniMax-H3 (MiniMax Community License); encoder base Qwen3-VL-32B; bf16 encoder packaging: Comfy-Org/MiniMax-H3.
• Abliteration: Arditi et al. 2024 (prior H3-encoder reference: ethanfel). Quantizer: convert-to-quant (silveroxides). Companion 4-bit DiT: lilcheaty; turbo LoRA: Gluttony10.
• License inherited from the base model.