Daruniawaesf 4K–16K Upscale Context-Tile Generator — V-1.0 Relase Krea2 — V-1.5 Anima

This ComfyUI workflow generates images at 4K, 8K, and 16K with staged context-aware diffusion refinement from text prompts or an input image. It provides controls for dimensions, seed, sampler, and pass steps, then saves lossless PNGs with workflow and generation metadata.

High-resolution passes use context-tile diffusion refinement. With 16K Memory Saver on, the workflow refines through 8K before one capped RTX VSR pass reaches 16K.

The Anima V-1.5 version adds HR Detail Strength and I2I Input Framing controls, Render Target selection, Tile Compute profiles, adaptive previews, and expanded render and PNG metadata.

PNG compression changes file size and encoding time, while the image remains lossless; the front-panel preview is reduced for 8K and 16K images.

Generation metadata records the selected checkpoint or Krea2 UNET, HR conditioning mode, refinement profile, sampling configuration, pass settings, and resolution pipeline.

This release uses ComfyUI's Classic/Legacy renderer; ComfyUI Nodes 2.0 beta is not required or recommended.

The release notes instruct updating existing custom nodes before use.

The Anima version lists anima-base-v1.0.safetensors, qwen_3_06b_base.safetensors, and qwen_image_vae.safetensors.

Required custom nodes listed are Context-Anchored Tile Refine, ComfyUI-DaSiWa-Nodes, rgthree-comfy, and ComfyUI-KJNodes.

To favor composition preservation, try Detail Only HR Conditioning with the Fidelity profile; these are the stated starting choices for that goal.

Sugestão · não verificado

For systems that cannot process native 16K diffusion, enable Q3 and 16K Memory Saver. For native 16K refinement, enable Q3 and turn Memory Saver off.

Sugestão · não verificado

SageAttention is optional and off by default; compatibility and performance gains depend on the GPU, PyTorch environment, model architecture, and attention implementation.

Native 16K processing can require very large amounts of system RAM and VRAM; requirements vary with checkpoint, aspect ratio, tile settings, and model architecture.

Semantic drift can still occur when denoise strength, prompt conditioning, or tile refinement settings exceed the selected model's capabilities.