Qwen Image 2.1 Morning Briefing 2026-11-10 | Top Pick: [Qwen 2.1] Sabikaze - grey-teal dry-brush anime style

This issue reviews 4 new Qwen 2.1 versions: 1 S-tier, 1 A-tier, and 2 B-tier. Top pick: Sabikaze, an anime illustration style LoRA featuring desaturated grey-teal tones, thick ink lines, and dry-brush textures. The author provides trigger word usage, steps/CFG/strength settings, and known issues. The A-tier entry is an oil painting style LoRA trained on classical paintings of Greek mythology from the 16–19th centuries, which includes trigger words and examples but no parameters. The Anima anime-to-realistic character sheet workflow has no visible examples and is listed as B-tier.

Zhenzhen's AI Radar · Qwen Image 2.1 · 2026-11-10 Morning Briefing

This issue features 4 new versions: Strongly Recommended 1 · Recommended 1 · Worth Watching 2 · Listed Only 0

Editor's Note This issue reviews 4 new Qwen 2.1 versions: 1 S-tier, 1 A-tier, and 2 B-tier. Top pick: Sabikaze, an anime illustration style LoRA featuring desaturated grey-teal tones, thick ink lines, and dry-brush textures. The author provides trigger word usage, steps/CFG/strength settings, and known issues. The A-tier entry is an oil painting style LoRA trained on classical paintings of Greek mythology from the 16–19th centuries, which includes trigger words and examples but no parameters. The Anima anime-to-realistic character sheet workflow has no visible examples and is listed as B-tier.

Strongly Recommended (1)

[S] [Qwen 2.1] Sabikaze - grey-teal dry-brush anime style · v1

LoRA · New Model · Author lepetitprince716355 · Released 11-10 00:02 · Downloads 22

Anime illustration style LoRA: Desaturated grey-blue/grey-teal tones with a small area of warm color accent, thick ink lines, flat shading blocks, and dry-brush textures on clothing and backgrounds.

A cold grey-teal base with a single warm accent, suitable for creating a series of illustrations with a unified color tone.

Images

Example 832x1216

Reasons for Recommendation

  • New model (the same-named LoRA also has Anima, MOSS Image, and Krea 2 versions; the author states they were all trained on the same batch of 57 self-generated images); serves anime-style character and scene illustrations with a quiet, melancholic atmosphere
  • Complete parameters: 24 steps, CFG 4, euler/simple, strength tested at 1.0, resolutions 832x1216 and 1216x832, loaded using LoraLoaderModelOnly
  • The author provided counter-examples: at 16 steps and CFG 1, the style emerges but the image breaks into color blocks; training used ai-toolkit, rank 32, 2000 steps, with the trigger word placed first in captions and painting style specified
  • 11 visible examples, no hidden examples

How to Use

  • Start the prompt with @z3-sabikaze., immediately followed by a sentence specifying the painting style (e.g., Anime-style digital illustration.), then describe the scene in complete sentences; if the painting style is not specified, the image may remain photorealistic
  • 24 steps, CFG 4, euler/simple, LoRA strength 1.0; the author's test environment was Qwen-Image 2.1 int8 ConvRot DiT + Qwen3-VL 8B text encoder

Risks and Limitations

  • Known issues listed by the author: small props may appear twice, hair color may drift if not specified in the prompt, requiring a seed change
  • Requires 24 steps + CFG 4; low-step acceleration settings cannot be used; subject to the Qwen Research License, the author cites Qwen's official statement that generated results belong to the user
  • Newly released, with few downloads and reviews so far

Trigger Word: @z3-sabikaze

Recommended (1)

[A] Classical Greek Mythology Paintings Style // LoRA // qwen-image-2.1 · v1.0

LoRA · New Model · Author bizs · Released 11-10 01:45 · Downloads 28

Classical oil painting style LoRA: Trained on classical paintings featuring Greek mythology from the 16–19th centuries, generating mythological oil paintings ranging from Renaissance and Baroque to Neoclassicism.

Images

Example 1376x912

Reasons for Recommendation

  • New model; serves classical oil painting style illustrations for mythological/historical themes. The author lists figures appearing in the dataset (Zeus, Hermes, Dionysus, Hercules, Achilles, Odysseus, etc.), creatures (Pegasus, Centaurs, Argus), and scenes (Judgment of Paris, Dionysian Revelry, Feast of the Gods, scenes from the Trojan War)
  • Trigger word gr33kmyth; the page provides checkpoint files for 1000 / 1200 / 1400 steps, with the main file being the 1200 step version
  • 4 visible examples; permissive license, sales allowed

How to Use

  • Add gr33kmyth to the prompt, then write the name of the mythological figure or scene; the author did not provide strength, steps, or CFG settings, nor did they explain the differences between the three checkpoints

Risks and Limitations

  • The model card only contains the dataset and character list, with no generation parameters
  • Only 4 visible examples; 1 sensitive-level examples are hidden

Trigger Word: gr33kmyth

Worth Watching (2)

[B] Marinette/LadyBug · Marinette/LadyBug (Qw2.1)

LoRA · Ported · Author WannedAI · Released 11-10 08:20 · Downloads 13

Character LoRA for Marinette, the protagonist of the animated series Miraculous Ladybug, containing two outfits in one file: casual wear and the Ladybug suit.

Images

Example 1152x2064

Reasons for Recommendation

  • Ported version (the same page also has Illustrious and Krea 2 versions); serves fan art illustrations requiring this character
  • Complete parameters: The author tested on Qwen Image 2.1 int8_convrot, with LoRA weight 0.8, CLIP strength 0.8, 25 steps, CFG 1.0, Euler/Simple
  • The author provided two complete trigger prompts: qmalad plus a casual wear description (black jacket, white T-shirt with floral pattern, pink pleated skirt), or a suit description (red bodysuit with black spots, red mask, red gloves and boots)

How to Use

  • Start the prompt with qmalad, followed by the author's casual wear or suit description to switch outfits; LoRA 0.8, CLIP 0.8, 25 steps, CFG 1.0, Euler/Simple

Risks and Limitations

  • The author explicitly states that generation is difficult to control on Qwen; copyrighted animated character
  • 9 visible examples, with 11 sensitive-level examples hidden (more hidden than visible)
  • Standard character LoRA with no new capabilities beyond the dual outfits, hence downgraded from S to B

Trigger Words: qmalad, blue hair, low pigtails, blue eyes, carefully lifting a fragile botanical print wearing her familiar open black jacket with patterned cuffs, white t-shirt with a pink-and-black floral graphic, and a plain solid pink pleated skirt with no pattern or embroidery on the skirt, face uncovered, qmalad, blue hair, low pigtails, blue eyes, wearing her recognizable fitted red superhero bodysuit with clean black polka dots, a red domino eye mask around both eyes, matching red gloves and boots

[B] Anima (anime) 2 Realistic character sheets · v1.0

Workflow · New Model · Author Raw_Fantasy · Released 11-10 02:21 · Downloads 18

ComfyUI workflow: Converts anime character images generated by the Anima model into realistic style character sheets (multi-angle turnarounds or expression grids). The author states that Qwen 2.1 analyzes the original image and writes prompts, which can serve as character references for video models like MiniMax H3 and LTX.

Images

Diagram 1448x1086

Reasons for Recommendation

  • New model (workflow, not new weights); serves the workflow of "first defining a character with an anime model, then converting it to a realistic character for video reference images"
  • The author claims it preserves approximately 90% of facial geometry and maintains consistency in facial features, clothing, and color scheme across the entire character sheet
  • Organized by groups and distinguished by color; to add view blocks, copy the example prompts from existing blocks

How to Use

  • After importing the JSON, use ComfyUI Manager's "install missing custom nodes" to add nodes; the author suggests adding MoreReal LoRA in the second subgraph with a strength not exceeding 0.6 (exceeding this distorts facial geometry); other models and parameters were not provided by the author

Risks and Limitations

  • All 5 examples are hidden, with no visible results
  • The author did not specify the source of the required Qwen 2.1 model files and Anima model
  • "Preserves approximately 90% of facial geometry" is the author's claim, with no comparison provided
  • The card image is an AI-generated diagram, not the actual model output; the author has no publicly available safe-for-work examples

Data: Civitai Public API (base model Qwen 2.1), collection window 11-09 20:31 → 11-10 09:01 (Beijing Time). Recommendation reasons are editor-reviewed; download counts only reflect attention.

Civitai

Explore the latest AI Radar updates