Zhenzhen's AI Radar · Qwen Image 2.1 · 2026-11-14 Evening Report
This issue features 2 new versions: Strongly Recommended 1 · Recommended 0 · Worth Watching 1 · Listed Only 0
Editor's Note This issue reviews 2 new versions of Qwen 2.1, all workflow updates, with 1 rated S and 1 rated B. The top recommendation is AusBoss's Turbo Generate + Edit workflow v1.3: the prompt box now includes input slots for images from the 2th to the 6th, allowing you to reference specific elements like "the jacket in image 2" for multi-image referencing; this works for both generation and editing. Zwaetschge's newly added Anime/3d 2 Real version lacks any release notes, so it can only be inferred from the name and filename that it converts anime / 3D to realistic style, resulting in an automatic downgrade from S to B.
Strongly Recommended (1)
[S] Qwen Image 2.1 Turbo Generate + Edit | 8 Steps, Auto Prompt · v1.3
Workflow · Version Update · Author: AusBoss · Published 11-14 14:02 · Downloads: 116
A workflow v1.3 using the official Qwen Image 2.1 Turbo for text-to-image and image editing: the prompt box now includes five image inputs from picture 2 to picture 6, allowing you to reference them in prompts by name, such as "image 2".
Images

Reasons for Recommendation
- Version update: Any image loader can connect to the new input slots; simply write "put the jacket from image 2 on her" in the prompt, applicable to both new generation and editing; the main loaded image is treated as image 1.
- The original three toggles are retained: Edit a picture (for editing, enabling the author's Consistency LoRA to maintain compositional alignment), Auto prompt (uses the loaded text encoder to expand short phrases into full prompts), and Upscale the result (uses the author's Restore LoRA for tiled upscaling).
- The author claims approximately 4 seconds for 2 MP on an RTX 5090; the entire model has received 3997 downloads and 11 comments, with this version including 1 videos and 1 images.
How to Use
- Requires AusBoss nodes 2.21.2 or higher (update if already installed) and ComfyUI 0.38 or higher; Models: Turbo int8 main model, Qwen3-VL 8B int8 text encoder, 2.1 VAE; editing and upscaling additionally require the author's two LoRAs.
- 8 steps are built-in with no Steps input box; the author states that setting a standard KSampler to 8 steps results in graininess; all three toggles default to off when loading the file.
Risks and Limitations
- Auto prompt only retains image numbering and does not analyze the additional images connected from the 2th to the 6th.
- Depends on the author's own node package; the license only permits in-site rental, not sale.
Worth Watching (1)
[B] Zwaetschge's Qwen Image 2.1 Turbo Workflows · Anime/3d 2 Real
Workflow · Version Update · Author: zwaetschge · Published 11-14 20:02 · Downloads: 6
The newly added "Anime/3d 2 Real" version in Zwaetschge's Qwen Image 2.1 Turbo workflow collection; based on the name and filename (Qwen21-Anime2Real.json), it appears to convert anime / 3D images to realistic style.
Images

Reasons for Recommendation
- Version update: The model card still contains the description from the previous version's 8 step text-to-image workflow—official 8 step sigma schedule, Euler sampler, CFG 1, using only built-in nodes, defaulting to a vertical 2 MP image.
- The same author's previous version had well-documented instructions, providing model files and directory placement; this version includes 4 visible examples.
How to Use
- Load Qwen21-Anime2Real.json; follow the author's provided schedule, Euler sampler, and CFG 1, do not add additional acceleration LoRAs; the input method and strength for realistic conversion are not specified by the author.
Risks and Limitations
- The author did not write update notes, so you must open the workflow yourself to see how this version works and what has changed.
- 2 sensitive-level examples are hidden; the license does not permit derivatives, only in-site rental.
Data: Civitai public API (base model Qwen 2.1), collection window 11-14 08:48 → 11-14 21:01 (Beijing Time). Recommendations are reviewed by editors; download counts only reflect attention levels.