darrellbest/Qwen-Image-2.1-PE-I2I-Heretic

Qwen-Image-2.1-PE-I2I Heretic (abliterated)

Not affiliated with or endorsed by Alibaba / Qwen. A community derivative of Qwen/Qwen-Image-2.1-PE-I2I, redistributed under the Qwen Research License (copy included as LICENSE). Non-commercial use only; commercial use needs a separate licence from Qwen.

The image-editing prompt rewriter for Qwen-Image-2.1, a fine-tuned Qwen3.5-VL 9B that turns a short edit instruction plus 1–N input images into a detailed English edit prompt, with its refusal behaviour removed by Heretic directional ablation. bf16, same shapes and parameter count as the source; nothing else was changed.

system_prompt.txt is included and required. It defines the output format. It is the unmodified file from the source repo.

Results

	Refusals	KL divergence
Original	100/100	0 (by definition)
This model (trial 998)	9/100	0.0368
Same, re-measured after export	9/100	0.0363

The second row is an independent evaluation of the exported weights with Heretic's evaluate_model. Refusals were measured on mlabonne/harmful_behaviors and KL divergence (damage to ordinary behaviour, first-token distributions) on mlabonne/harmless_alpaca, both Heretic's defaults, with Heretic's default system prompt.

With n = 100 the refusal count is noisy (σ ≈ 3 at this rate), so 6 vs 9 vs 11 are not meaningfully different. Trial 998 was chosen from the Pareto front for having the damage level of the widely used T2I Heretic (pottokao/Qwen-Image-2.1-PE-T2I-Heretic, KL 0.036):

Refusals	KL
6/100	0.0431	trial 691
9/100	0.0368	trial 998 (this model)
11/100	0.0330	trial 960
16/100	0.0307	trial 957
20/100	0.0265	trial 975
31/100	0.0194	trial 682

Does it still rewrite edits?

Checked through a real editing app's prompt-enhancer path (the official pe_core output contract, enable_thinking, Qwen's sampling settings) on three edit instructions: a plain one ("make the sky a dramatic sunset") and two borderline ones (horror-film fake blood; a prop handgun on a table). Every answer parsed, and the rewrites were as detailed and specific as the original's.

An honest caveat: the original model did not refuse those image edits either. Like the T2I rewriter, it is relaxed about edgy image subjects and near-total in refusing chat-style harmful instructions, which is what the 100/100 baseline measures. The ablation mainly changes the latter.

NVFP4 build for vLLM on Blackwell: darrellbest/Qwen-Image-2.1-PE-I2I-Heretic-NVFP4. Companion for text-to-image: darrellbest/Qwen-Image-2.1-PE-T2I-Heretic.

Reproduce

• Heretic 3521f86 (main, 2026-09-21), default config except n_trials = 1000 (200, then 800 more on the same study) and export_strategy = "merge". transformers 5.17.0, torch 2.11.0+cu130, one RTX PRO 6000 Blackwell.
• Trial 998 parameters: direction_scope = global, direction_index = 19.522, attn.o_proj: max_weight 0.977 at 19.045, min_weight 0.935, min_weight_distance 18.014, mlp.down_proj: max_weight 1.456 at 22.104, min_weight 0.831, min_weight_distance 18.551.

Use

Exactly like the original: load with AutoModelForImageTextToText / AutoProcessor and use system_prompt.txt as the system prompt.

The family

Model	Format	Use it with
PE-T2I-Heretic	bf16 safetensors	transformers / diffusers
PE-T2I-Heretic-GGUF	GGUF BF16 / Q8_0 / Q4_K_M	llama.cpp, Ollama
PE-T2I-Heretic-NVFP4	NVFP4 (compressed-tensors)	vLLM on Blackwell
PE-I2I-Heretic	bf16 safetensors	transformers / diffusers
PE-I2I-Heretic-GGUF	GGUF BF16 / Q8_0 / Q4_K_M (+ mmproj)	llama.cpp, Ollama
PE-I2I-Heretic-NVFP4	NVFP4 (compressed-tensors)	vLLM on Blackwell

T2I rewrites a short request into a detailed prompt for new images; I2I rewrites an edit instruction, reading the image being edited. Both are prompt rewriters for Qwen-Image-2.1, not image generators.