Builders looking to squeeze more out of anime-oriented image generation in ComfyUI now have a comprehensive toolkit. The Anima Aesthetic 1.1 release includes ten distinct JSON workflow graphs, ranging from simple baselines to complex multi-stage pipelines. These aren't just theoretical models; each graph is a saved configuration ready for import, covering everything from text-to-image generation to tiled upscaling and facial repair.

The Spectrum of Complexity

The workflows demonstrate a wide range of node counts, signaling different use cases for developers. At the lower end, the Text-to-Image Baseline uses just 17 nodes with a 30-step er_sde sampler at CFG 4, while the DMDX 4-Step Speed Run strips it down to 19 nodes for rapid iteration using the anima-turbo-lora-v0.2 and dmdx_student_final_comfy. Conversely, the Image-to-Image + ControlNet + Face/Hand Repair workflow balloons to 52 nodes, incorporating six UltralyticsDetectorProvider nodes and four FaceDetailer passes to ensure structural integrity in the output.

Precision and Repair Pipelines

For those needing exact control, the ControlNet Precision Control workflow integrates pose, depth, edge, and line-art branches into a single graph. It utilizes the anima-base-202605-v1.0 UNET with a two-pass denoise strategy (1.0 then 0.55). Meanwhile, the 1.5x Upscale workflow leverages the Qwen-Image HDR VAE and qwen_3_06b_base text encoder, running two ClownsharK Beta passes to refine details without excessive compute overhead.

Critical Implementation Notes

A crucial caveat for builders: these JSON files do not bundle the model weights. Every workflow references specific files like anima-base-202605-v1.0, waiANIMA_v10, and various LoRAs that must already exist in your local environment. Furthermore, the source explicitly states that no timing or quality benchmarks were run for this release, meaning performance metrics are anecdotal and dependent on your specific hardware and prompt seeds.

Key Takeaways

  • Ten distinct ComfyUI graphs provided for Anima Aesthetic 1.1, covering baselines, ControlNet, and repair tasks.
  • Node counts vary significantly, from 17 nodes for baselines to 60 nodes for complex text-to-image repair pipelines.
  • Workflows rely on specific external models (e.g., Qwen-Image HDR VAE, UltralyticsDetectorProvider) that are not included.
  • No official benchmarks for speed or quality were provided in the release notes.

The Bottom Line

This is a solid, practical addition for ComfyUI power users, but don't expect a plug-and-play experience without ensuring your model directory is fully stocked first.