Light Novel style book illustrations with anima-preview2
A viral workflow combines Qwen3-VL, Flux-Klein-9b, and manual editing to generate chapter-by-chapter illustrations.
A developer has demonstrated a sophisticated, multi-model AI pipeline that automates the creation of professional light novel illustrations directly from chapter text. The viral workflow, shared by RobertTetris, uses huihui_ai's Qwen3-VL-8b model to analyze each chapter, identify the most iconic moment, and generate detailed image prompts. These prompts are then fed into the anima-preview2 image generation model using a standard workflow with specific settings: cfg=4.0 and 30 steps. The system maintains character consistency by referencing a character sheet and cover art, producing images that align closely with the narrative.
Post-generation refinement is handled through a combination of automated and manual techniques. The Flux-Klein-9b model is used within ComfyUI for initial edits, often guided by a sample character image. Final touches are applied manually in Krita, utilizing Krita-AI-Diffusion with various models at lower weights for subtle refinements. The developer focused edits primarily on maintaining hairstyle and t-shirt consistency across illustrations, with occasional fixes for anatomical details like finger counts. The result is a set of illustrations with high textual accuracy for a story originally published on Royal Road, showcasing a replicable method for authors and illustrators to scale professional-grade artwork for serialized content.
- Uses Qwen3-VL-8b to analyze chapter text and generate prompts for key moments
- Leverages anima-preview2 with cfg=4.0 and 30 steps for initial image generation
- Employs Flux-Klein-9b and manual Krita editing for consistency and detail refinement
Why It Matters
This pipeline demonstrates how authors can automate high-quality, consistent illustration at scale, reducing production time and cost for serialized content.