Model Download Link: https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/tree/main

V1: Uses Qwen-Image-Edit-2509 and 4-step Lightning v2.0. Contains a small amount of NSFW LoRAs, making it a general-purpose model suitable for both SFW and NSFW scenarios. Recommended: sa_solver/beta, but euler_a/beta and er_sde/beta also deliver decent results.

V2: Now blends Qwen-Image-Edit accelerators, merging 8-step and 4-step into one step. Major adjustments were made to NSFW LoRAs for more comprehensive coverage across SFW and NSFW scenarios. Highly recommended: sa_solver/simple.

V3: Uses the new Qwen-Image-Edit Lightning LoRA for better results. The NSFW LoRA mix was heavily adjusted, removing lower-quality LoRAs and adding higher-quality ones. Highly recommended: sa_solver/beta.

V4: Blends multiple Qwen Edit and base Qwen accelerators, which I think works better. Added some LoRA skin corrections. 4–5 steps: use sa_solver/simple, lcm/beta, or euler_a/beta; 6–8 steps: use only lcm/beta or euler_a/beta.

V5: NSFW and SFW use cases were interfering with each other too much, so I split them up to better serve their respective scenarios. v5.2 updated the "snofs" and "qwen4play" NSFW LoRAs + Meta4; v5.3 added "Qwen Image NSFW Adv." by fok3827. SFW: lcm/beta or er_sde/beta are generally recommended. NSFW: lcm/normal is recommended. Prompting "professional digital photography" helps reduce the plasticky look.

V6: Attempted to use valiantcat/Qwen-Image-Edit-MeiTu and chestnutlzj/Edit-R1-Qwen-Image-Edit-2509 as base models. However, the merge failed. It seems using them as LoRAs might work better; I need further testing to find a working approach. Please stick to v5 until a new version is released.

V7: Included valiantcat/Qwen-Image-Edit-MeiTu and chestnutlzj/Edit-R1-Qwen-Image-Edit-2509 as LoRA files. Adjusted accelerators and NSFW LoRA files (v7.1 contains more NSFW content). Results seem much better. Suggested: 4–6 steps with lcm/sgm_uniform, 7–8 steps with lcm/normal.

V8: Loads FP32 LoRA data using BF16, scaling down to FP8 only upon saving. This seems to help resolve grid issues and improve quality. Adjusted acceleration ratio. Major adjustments to LoRA data (plus new SNOFS). Suggested: 4–6 steps with euler_a/beta, 7–8 steps with lcm/normal.

V9: Okay, I lied. The "rebalance" and "smartphone photorealism" LoRAs do help generate images suitable for all ages (SFW) and adult content (NSFW). If you don't want these LoRAs integrated (e.g., for making anime or cartoons), use the "Lite" version. Also, a typo in the accelerator settings in V8 was fixed in V9. Adjusted NSFW LoRAs and significantly reduced their runtime overhead, which should help improve consistency. Recommended: 4–6 steps with euler_a/beta. Generally, using sgm_normal or a normal scheduler with higher step counts yields better results.

V10: A mix between v5 and v9. MeiTu and Edit-R1 have been removed. I kept the strength of the "rebalance" and "smartphone" LoRAs at half, which I think helps improve skin texture, diversity, and composition. NSFW LoRAs are very similar to v5.3 (but updated with snofs v1.2). v10.4 NSFW was adjusted to improve character consistency and penis rendering. Strongly recommended: euler/beta for 4–8 steps, but euler_a/sgm_uniform is recommended for NSFW v10.2 and above.

V11: Adjusted NSFW LoRAs, reducing the quantity used in favor of more compatible ones. Spread out the "realism" LoRAs into multiple lower-strength LoRAs. Recommended: euler/beta for both NSFW and SFW, but feel free to experiment with other setups!

V12: Got impatient waiting for Qwen Edit 2511 (or 2512?), so I went looking for more LoRAs and tweaks to reduce the "plasticky" look. euler/sgm_uniform works for SFW, er_sde/sgm_uniform works for NSFW, though experimenting with different setups is beneficial in this case.

V13: Tweaked built-in LoRAs to reduce grid lines and improve character consistency (while avoiding a plasticky look). Recommended: er_sde/beta.

V14: Trimmed down LoRAs that might hurt character consistency (hopefully reducing the "plasticky" look as well). Updated to the new SNOFS v1.3 for NSFW content, and trimmed underperforming NSFW LoRAs. Recommended: er_sde/beta.

V14.1: Added "InSubject" LoRA to both SFW and NSFW merges to improve character consistency. Otherwise largely identical to v14. Recommended: er_sde/beta.

V15: Now using Qwen-Edit-2511! Minor tweaks to LoRAs. Skipped the NSFW merge since the base merge didn't turn out well. Recommended: er_sde/beta, but feel free to experiment.

V16: Noticed performance degradation from some LoRAs in previous versions (mainly "realism" LoRAs) that weren't needed or compatible with 2511, so trimmed them down. Further tweaked NSFW LoRAs (added "Sex-tacular"). Recommended: er_sde/beta.

V17: Merged 2509 and 2511 to fix contrast issues and improve 2511's compatibility with LoRAs while maintaining character consistency. Highly recommended: euler_ancestral/beta.

V18: Increased 2511 weighting and reduced 2509 weighting. Used a different merging method, which I think resolves issues previously encountered when processing large numbers of input images. Recommended: euler_ancestral/beta.

V19: Added Lightning Edit 2511 8-step version (4–8 steps still recommended). Also added a new NSFW LoRA (GNASS for Qwen 2512), which performs well in the NSFW merge. Recommended: er_sde/beta or euler_ancestral/beta.

V20: Reverted to 100% Qwen Edit 2511. Adjusted accelerators, hoping for better results without blending in 2509. Added a small amount of "best face swap". Adjusted realism LoRAs and made major adjustments to NSFW LoRAs. Recommended: euler_ancestral/beta.

V21: Removed old "realism" LoRAs and added 2511 "anything2real" and "anime2real" LoRAs to improve skin texture and overall editing capabilities. Adjusted NSFW LoRA weights. If you are only interested in flat 2D anime, v20 or v19 might suit you better (as this version leans more toward photorealism). Recommended: euler_a/beta.

V22: Removed the "anything2real" and "anime2real" add-ons again, as they ruined style consistency. Added a small amount of JibMix Skin and qwen-skin-edit v1.1 to address the "plasticky" issue. Major adjustments to NSFW LoRA weights. Recommended: euler_ancestral/beta.