Zhenzhen's AI Radar · Qwen Image 2.1 · 2026-11-04 Evening Report
This issue features 4 new versions: Strongly Recommended 1 · Recommended 2 · Worth Watching 1 · Listed Only 0
Editor's Note This issue reviews 4 new versions of Qwen 2.1. The top recommendation is a character sheet workflow: from one portrait photo, it generates front, three-quarter, side, and back views. It can add head close-ups or four expressions, includes a fast mode and 2x upscaling, and provides comprehensive details on dependencies, model paths, and dimensions. Tier A features the same author's "click-to-select" inpainting workflow v1.1 (with new copy-from-other-image functionality) and an edit workflow for changing poses while preserving faces; the Edit LoRA with tail addition has only two sentences of description, so it is rated B.
Strongly Recommended (1)
[S] Qwen Image 2.1 Character Sheet | Turnaround, Close-Ups + Expressions From One Photo · v1.0
Workflow · New Model · Author AusBoss · Published 11-04 12:55 · Downloads 141
Qwen Image 2.1 character sheet workflow: Load one portrait photo to generate a side-by-side character sheet with front, three-quarter, side, and back views of the same person. It can add a row of head close-ups or four expressions.
The same author's H3 version is also in use. The Qwen version is suitable for quickly generating character sheets first, then feeding them into H3 as reference images for video generation.
Reasons for Recommendation
- New model (workflow): The author adapted their MiniMax H3 character sheet workflow to a Qwen 2.1 version. The author claims it is faster and smaller, taking about 7 seconds per image.
- Serves character design and consistent assets: Outputs one image at 1344 × 768, with each of the four views at 336 × 768. Parts missing from the photo (such as legs in a half-body shot) can be specified in the input box to be completed.
- Comprehensive toggles: Head close-up row (image becomes 1344 × 1104), four expressions, Viggle Turbo fast mode, SeedVR2 2x upscaling, individual saving of each view, and side-by-side original image display.
- Easy to replicate: Lists minimum ComfyUI and node versions, every model filename, and placement directory. Free; includes 4 images + 1 video examples; 141 downloads (indicating popularity only).
How to Use
- Requires ComfyUI 0.38 or higher, AusBoss nodes 2.15.0 or higher. Models used are the standard qwen_image_2.1_int8_convrot (not Turbo), qwen3vl_8b text encoder, and 2.1 VAE.
- Load a clear photo of a person. Background can be specified in the input box (leave blank for white background). Clothing not visible in the photo should be written into "What the photo doesn't show".
- For speed, enable Fast mode: Loads Qwen-Image-2.1-viggle-turbo-v0.3-6step-lora-r128, using 6 steps instead of 25. First run requires loading approximately 17 GB of models.
Risks and Limitations
- The back view is guessed by the model based on the front. When expanding from half-body to full-body, small details (such as a shirt under a jacket) may be lost and need to be specified in the supplementary box.
- Both Qwen Image 2.1 and Viggle Turbo are under the Qwen Research License; commercial use requires separate authorization. The workflow license only permits on-site rental.
- Speed figures were tested by the author on an RTX 5090. This entry has not been run by us.
Recommended (2)
[A] Same Face, New Pose — Qwen-Image 2.1 Edit Workflow · v1
Workflow · New Model · Author TP_AI_63 · Published 11-04 19:08 · Downloads 32
Qwen 2.1 Edit workflow for changing poses while preserving faces: Provide a character reference image to generate new pose images of the same character, retaining face, hairstyle, and skin tone without needing LoRA or training.
Reasons for Recommendation
- New model (workflow): Includes 12 preset poses (standing, sitting, walking, half-body, front/side/back turns, close-ups, etc.) and a custom pose text box. Two clothing modes (keep original clothing / casual clothing).
- Serves consistent character image generation: Batch generate different poses for your own characters. Built-in SeedVR2 2x upscaling can be disabled with one click.
- Lists model files, placement directories, download links, and custom nodes. The author admits only some poses were checked. Free; 9 visible examples; 32 downloads (indicating popularity only).
How to Use
- Drag the JSON into ComfyUI. In Group 1, select the reference image. In Group 2, set clothing mode to 1 or 2. In Group 3, set pose from 1 to 12. In Group 4, write a sentence describing the scene and lighting.
- Models: qwen_image_2.1_int8_convrot, qwen3vl_8b_int8_convrot, 2.1 VAE. Use SeedVR2 for upscaling. Install missing nodes via Manager.
Risks and Limitations
- The author tested with only 1 characters and 1 reference images. Poses 5, 7, 8, 9, and 10 were not tested. With landscape reference images, pose 12 cuts off at the thighs.
- The author states that enabling SeedVR2 peaks at approximately 15 GB VRAM; disable upscaling if VRAM is insufficient.
- The author requires usage only for self-generated adult characters or authorized photos. Qwen models are under research license. This entry has not been run by us.
[A] Qwen Image 2.1 Inpaint | Click to Mask · v1.1
Workflow · Version Update · Author AusBoss · Published 11-04 09:18 · Downloads 170
The "click-to-select" Qwen 2.1 inpainting workflow v1.1: Click objects in the image and describe changes in one sentence to repaint only the selected area. This version adds functionality to copy elements from other images.
Reasons for Recommendation
- Version update: You can reference images from 2 to 6; in the prompt, write “replace the jacket with the jacket from image 2”; when no images are referenced, it behaves exactly like v1.0
- Precise local editing service: SAM 3 finds contours based on clicks, preserving original pixels outside the selection; the author states that selected parts are processed at approximately 1.5 million pixels and then pasted back, ensuring detail even for small objects
- Dependencies, model files, and directories are fully documented, with a YouTube demo included; free; this version has 170 downloads, 1790 cumulative model downloads, and 9 comments (reflecting only attention levels)
How to Use
- Requires ComfyUI 0.38 or higher, AusBoss nodes 2.12.0 or higher (existing users should update via Manager first). Models: qwen_image_2.1_int8_convrot, qwen3vl_8b_int8_convrot, 2.1 VAE, sam3.1_multiplex_fp16.
- Load image → Click objects in "Mask by Touch" → Describe changes in one sentence in "What to change" → Run. To borrow from other images, add image loading nodes yourself and connect them to picture 2 and beyond.
- Optional Fast mode: Viggle Turbo LoRA, using 8 steps instead of 25. The author claims this halves the time required.
Risks and Limitations
- Visible examples include only 1 videos (another 1 sensitive-level examples are hidden).
- Qwen models are under research license; commercial use requires separate authorization. This entry has not been run by us.
Worth Watching (1)
[B] Tail - Qwen Edit · Qwen 2.1 v1.0
LoRA · New Model · Author 2_as_1 · Published 11-04 16:46 · Downloads 38
A "add tail" LoRA for Qwen 2.1 Edit: Adds a tail to the character in the input image. Drawing a red line can specify the tail's direction.
Reasons for Recommendation
- New model: Description is only two sentences—adds a tail; draw a red line on the input image, and the tail will generate along the red line.
- Serves character modification in image editing. Red line guidance makes tail position controllable.
- 2 visible examples; free with permissive license. The same author released an H3 version around the same time.
How to Use
- Load tail-qwen_image_21_edit-v7.e016.safetensors. Write "add a tail" in the edit instruction. To control direction, draw a red line on the input image.
- Strength and step count were not provided by the author.
Risks and Limitations
- Description is extremely short; another 4 sensitive-level examples are hidden.
- Requirements for drawing the red line (thickness, color) were not specified by the author. This entry has not been run by us.
Trigger Words: add a tail
Data: Civitai public API (base model Qwen 2.1), inclusion window 11-04 09:00 → 11-04 21:00 (Beijing Time), backfill period. Recommendation reasons reviewed by editors; download counts indicate popularity only.