Zhenzhen's AI Radar · Qwen Image 2.1 · 2026-11-11 Morning Report
This issue features 4 new versions: Strongly Recommended 0 · Recommended 0 · Worth Watching 1 · Listed Only 3
Editor's Note This issue reviewed 5 new Qwen 2.1 versions, with 0 rated S/A, 1 rated B, and 3 rated C; 1 were excluded. The most practical is MosaicMate: an editing LoRA that adds "semi-transparent mosaic removal" capability to Qwen-Image 2.1. Parameters are fully provided, but the author's tags explicitly state "decensor," and there are only 1 example images, so it was downgraded from A to B and is not a top recommendation. The rest are three character/role LoRAs (listed as C); Moss-yul Nangong He's files and descriptions are for the MOSS version but were labeled as Qwen 2.1, so they have been excluded.
Worth Watching (1)
[B] Qwen Image 2.1 MosaicMate LoRA | Remove semi-transparent mosaic · v1.0
LoRA · New Model · Author SchwiAI · Published 11-10 22:57 · Downloads 60
Image editing LoRA: Enables Qwen-Image 2.1 to more accurately remove semi-transparent mosaic overlays from images, primarily relying on residual textures and structures beneath the overlay to restore details rather than guessing based on priors.
Downgraded from automatic A to B: Parameters are complete and the use case is clear, but there are only 1 example images, and the primary purpose is censor removal, so it is not a top recommendation.
Images

Reasons for Recommendation
- New model: The author states that Qwen-Image 2.1 itself can already remove some mosaics; this adapter improves restoration accuracy for semi-transparent mosaics (with opacity up to 0.9)
- Serves repair-type tasks in image editing: Input an image with a semi-transparent mosaic, provide an editing instruction, and output the cleaned image
- Parameters fully provided: Strength 1.0, CFG 1.0, euler / simple; Base 10-20 steps, Turbo 4-8 steps, no trigger word required
How to Use
- Load the LoRA with strength 1.0, CFG 1.0, and sampler euler / simple; use 10-20 steps for Base and 4-8 steps for Turbo
- Write the editing instruction as "Remove the semi-transparent mosaic"; no trigger word is needed
- The author recommends cropping the image to the area containing the mosaic before processing, to save time and avoid affecting other regions
Risks and Limitations
- The author's tags include "decensor," indicating a bias toward censor removal; it should not be used on mosaics in photos of others intended for privacy protection (such as faces or license plates)
- There are only 1 example images; comparison results must be viewed on the author's other pages
- The author states that pixel-level restoration is not guaranteed, and severely occluded areas may not be recoverable; it targets only semi-transparent overlays, as opaque mosaics were not included in the training scope
Listed Only (3)
- Valeria | MOSS / Krea 2 /Qwen 2.1 · v1.0 Qwen 2.1 — LoRA: A Qwen 2.1 port of the original fictional female character Valeria (approximately 40 years old, mature appearance with glasses), used for generating portraits with a consistent face.

- 000_yul_Nangong He (Wan) Qwen2.1 · 000_yul_Nangong He (Wan) Qwen2.1 — LoRA: A BJD doll-textured ancient-style female character LoRA (Nangong He): blue and white ancient-style long dress, silver hair accessories, triggered by a full Chinese prompt.

- Witches: Yennefer cosplay | MOSS / Krea 2 /Qwen 2.1 · v1.0 Qwen 2.1 — LoRA: A Qwen 2.1 version of the Yennefer cosplay portrait LoRA: dresses the author's original character Kira as Yennefer from The Witcher (purple eyes, black curly hair, black velvet long dress, and star-shaped necklace).

Data: Civitai public API (base model Qwen 2.1), inclusion window 11-10 20:31 → 11-11 13:03 (Beijing Time). Recommendation reasons have been reviewed by editors; download counts reflect attention only.