Qwen Image 2.1 Morning Report 2026-10-30

This issue reviewed 5 new versions of Qwen 2.1, with no S / A entries. The top-ranked item is a three-stage workflow: first generating images with Krea 2, then editing them using the Krea 2 identity editing LoRA, and finally handing them over to Qwen 2.1 for editing. The dependency list is comprehensive, but it lacks usage steps and parameters, and the models are often tens of GB in size, so it was downgraded from A to B. The two editing-style LoRAs, Mima Cake and SweetCreamCake, have trigger phrases that can be used directly, but their model cards are only one line long, so they remain at B. The two standard character LoRAs, White Dragon Girl and Whale Girl, were trained using GPT Image 2.5 for image editing and lack usage instructions, so they were downgraded to C.

Zhenzhen's AI Radar · Qwen Image 2.1 · 2026-10-30 Morning Report

This issue features 5 new versions: Strongly Recommended 0 · Recommended 0 · Worth Watching 3 · Listed Only 2

Editor's Note This issue reviewed 5 new versions of Qwen 2.1, with no S / A entries. The top-ranked item is a three-stage workflow: first generating images with Krea 2, then editing them using the Krea 2 identity editing LoRA, and finally handing them over to Qwen 2.1 for editing. The dependency list is comprehensive, but it lacks usage steps and parameters, and the models are often tens of GB in size, so it was downgraded from A to B. The two editing-style LoRAs, Mima Cake and SweetCreamCake, have trigger phrases that can be used directly, but their model cards are only one line long, so they remain at B. The two standard character LoRAs, White Dragon Girl and Whale Girl, were trained using GPT Image 2.5 for image editing and lack usage instructions, so they were downgraded to C.

Worth Watching (3)

[B] Krea 2 + Krea 2 Edit + Qwen Image 2.1 Edit v 1.0 · v1.0

Workflow · New Model · Author yarcomics514 · Published 10-30 02:39 · Downloads 58

A ComfyUI workflow JSON that chains Krea 2 image generation, Krea 2 identity editing, and Qwen 2.1 editing into three steps; it is not new weights.

Downgraded from A to B: Dependencies are clearly documented, but it is only a workflow, lacks operational instructions, and has a high barrier to entry (dozens of large models).

Images

Example 1629x1073

Reasons for Recommendation

  • New model: The three steps involve generating images with Krea 2, editing images while preserving character identity using the Krea 2 Identity Edit V2 LoRA, and then performing edits with the Qwen 2.1 model; the name and base model are explicitly stated as Qwen Image 2.1, confirming it belongs to this section
  • The author lists dependencies comprehensively: 8 custom nodes (rgthree-comfy for flow switching, ComfyUI-VAE-Utils for tiled VAE decoding) and a complete model directory structure
  • Serves image editing workflows requiring "generate first, then edit while maintaining character consistency"; free, license allows output, rental, and sale of fusions

How to Use

  • Download krea2Krea2EditQwenImage2_v10.json and import it into ComfyUI, use Manager to install ComfyUI-Easy-Use, ComfyUI-Custom-Scripts, ComfyUI ArtVenture, ComfyUI-post-processing-nodes, Pixaroma, ComfyUI-OreX, rgthree-comfy, and ComfyUI-VAE-Utils
  • The Qwen 2.1 part uses qwenvy21_v10.safetensors (int8, 6.76 GB), the Qwen-Image-2.1-viggle-turbo-v0.3-6step-lora-r256 acceleration LoRA, the texture_fix_vae_for_qwen_image_2_1 VAE, and the text encoder qwen3vl_8b_fp8_scaled; the author did not provide steps, CFG, or how to switch between steps

Risks and Limitations

  • Heavy dependencies: Each Krea 2 base model is around 12 GB, plus the text encoder, VAE, and multiple LoRAs; the Qwen part uses a community fine-tuned base model rather than the official base model
  • Among the Krea 2 base models listed by the author, two are marked as sensitive content, and another text encoder is the Heretic uncensored version; pay attention when selecting models
  • The model card only contains a dependency list and a one-sentence workflow introduction, with no operational steps; 3 visible examples are available, with 4 sensitive-level examples hidden; this item was not run

[B] Mima Cake Style · QE21 | v1

LoRA · Port · Author spanking_cutie · Published 10-30 05:46 · Downloads 22

An editing-style LoRA that converts input images into the "mimacake" anime style, Qwen 2.1 version; triggered by a fixed instruction.

Remains at B: The trigger phrase can be used directly and the license is permissive, but the description is only one line, and the effect can only be judged from 3 examples.

Images

Example 1024x1024

Reasons for Recommendation

  • Port: Previous versions existed for Krea 2 and Anima (those two versions used @mimacake as the trigger word); the Qwen 2.1 version changed to an editing instruction-style trigger phrase, with other differences not specified by the author; the entire model has accumulated 948 downloads and 86 likes (representing attention only)
  • Serves image-to-image / image editing: Converts existing images into this anime style
  • Free, license allows output, rental, sale, and sale of fusions

How to Use

  • Load mimacakeQ21_000001750.safetensors (BF16) in the Qwen 2.1 editing workflow, input the original image, and write the instruction "transform the image into a mimacake anime style"
  • The author only specified the maximum resolution 1024x1024; strength, steps, and CFG were not provided by the author

Risks and Limitations

  • The complete model card is also just one line: "Max Res 1024x1024", with no explanation of style characteristics or training material sources
  • Only 3 visible examples are available, with 7 sensitive-level examples hidden
  • The filename includes the training step, suggesting it is an intermediate checkpoint; the author did not specify if it is the final version

Trigger Word: transform the image into a mimacake anime style

[B] SweetCreamCake Style · QE21 | v1

LoRA · Port · Author spanking_cutie · Published 10-30 03:20 · Downloads 18

An editing-style LoRA that converts input images into the "sweetcreamcake" anime style, Qwen 2.1 version; triggered by a fixed instruction.

Remains at B: Same author and approach as Mima Cake, the trigger phrase can be used directly, but the description is only one line.

Images

Example 1440x1440

Reasons for Recommendation

  • Port: Previous versions existed for Illustrious, Anima, and Krea 2 (the latter two versions used @sweetcreamcake as the trigger word); the Qwen 2.1 version changed to an editing instruction-style trigger phrase, with other differences not specified by the author; the entire model has accumulated 1532 downloads and 144 likes (representing attention only)
  • Serves image-to-image / image editing: Converts existing images into this anime style
  • Free, license allows output, rental, sale, and sale of fusions; includes 5 visible examples

How to Use

  • Load sweetcreamcakeQ21_000002000.safetensors (BF16) in the Qwen 2.1 editing workflow, input the original image, and write the instruction "transform the image into a sweetcreamcake anime style"
  • The author only specified the maximum resolution 1024x1024; strength, steps, and CFG were not provided by the author

Risks and Limitations

  • The complete model card is also just one line: "Max Res 1024x1024", with no explanation of style characteristics or training material sources
  • 15 sensitive-level examples are hidden, far more than the 5 visible examples
  • The filename includes the training step, suggesting it is an intermediate checkpoint; the author did not specify if it is the final version

Trigger Word: transform the image into a sweetcreamcake anime style

Listed Only (2)

Data: Civitai public API (base model Qwen 2.1), inclusion window 10-29 20:31 → 10-30 09:01 (Beijing Time). Recommendation reasons have been reviewed by editors; download counts represent attention only.

Civitai

Explore the latest AI Radar updates