Developer Notes: After ~23 iterations of this "all in one", I'm winding down providing updates. v19 and v23 are probably the best versions. I feel this project has peaked as new image edit models are being released. Thank you for all the feedback and help!
V23 (latest) : Tweaks to skin and realism LORAs (removed JibMix which seemed to cause oddities). Reduced a "plastic" NSFW LORA that was making realistic results more difficult in that merge. This one was feeling pretty good. euler_ancestral/beta recommended
V22: Removed "anything2real" and "anime2real" which were stomping styles and causing inconsistencies. Added small amounts of JibMix Skin and qwen-skin-edit v1.1 to combat 'plastic' instead. Significant tweaks to NSFW LORA weights again. euler_ancestral/beta recommended
V21: Removed older "realism" LORAs and added new 2511 "anything2real" and "anime2real" LORAs in hopes of improving skin texture and overall edit functionality. Tweaks to NSFW LORA weights. euler_a/beta recommended
V20 : Returning to 100% Qwen Edit 2511. Tweaks to accelerators to hopefully give better results without 2509 mixed in. Added a minor amount of "BestFaceSwap". Tweaks to realism LORAs and significant tweaks to NSFW LORAs. euler_ancestral/beta recommended
V19 : New Lightning Edit 2511 8-step mixed in (still recommend 4-8 steps). Also a new NSFW LORA (GNASS for Qwen 2512) that worked quite well in the NSFW merge. er_sde/beta or euler_ancestral/beta recommended.
V18: More 2511, a bit less 2509. Different merging method that I think corrects an issue I was having with many input images. euler_ancestral/beta recommended
V17: Merged 2509 and 2511 together with the goal of correcting contrast issues and LORA compatibility with 2511 while maintaining character consistency. euler_ancestral/beta highly recommended.
TA Notes:
When remixing a post TA will take you to img2img , this is a site bug. The Qwen Edit model shall be used under 'Edit' section in creation tab
For an Edit output you need to provide a reference image. Showcases used here uses landscape 3DCG images as the base image. See this article for example use cases with QWEN edit: https://huggingface.co/blog/MonsterMMORPG/qwen-image-edit-full-tutorial-26-different-demo
The QWEN Edit model affects the latent representation of the image , a numerical vector. Since the latent representation of an image is abstract , you can map a landscape 1200x600 to a 1024x1024 square or vice versa. Very useful for prepping LoRa images! See example here https://tensor.art/articles/946881918559878715
The CFG parameter sets the ratio between conditional and unconditional prompting. The formula is X = x_cond • CFG + (1-CFG) * x_uncond
At low CFG , CFG < 1 , the reference image (unconditional prompt) , guides the output . At higher CFG >= 1 , the text prompt (conditional prompt) guides the output.
Repo page:
https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO
This is an fp8 checkpoint (dtype is fp8_e4m3fn)
List of LoRas merged into this checkpoint:
Qwen_Snofs_1_3.safetensors
yky_000032000.safetensors
Meta4.safetensors
Qwen S3x-tacular V1.safetensors
qwen_MCNL_v1.0.safetensors
qwen-image_n5fw_adv_v1.0.safetensors
Qwen4Play_v2.safetensors
Qwen-Image-Edit-2509-Lightning-8steps-V1.0-fp32.safetensors
Qwen-Image-Edit-2511-Lightning-4steps-V1.0-fp32.safetensors
Qwen-Image-Edit-2509-Lightning-4steps-V1.0-fp32.safetensors
Rebalance_v1_lora_r32.safetensors
Real-life-Simulator.safetensors
Qwen-Image_SmartphoneSnapshotPhotoReality_v4.safetensors
iphone.safetensors
qwen-boreal-portraits-portraits-high-rank.safetensors
InSubject-0.5.safetensors
//-----//
Developers notes:
Merge of accelerators, VAE and CLIP to allow for easy and fast Qwen Image Edit (and text to image) support. Use a "Load Checkpoint" node. 1 CFG, 4 step. Use the "TextEncodeQwenImageEditPlus" node for input images (which are optional) and prompt. Provide no images to just do pure text to image. FP8 precision.