RunningHub workflow page: https://www.runninghub.ai/zh-cn/post/2089343542203346946?inviteCode=rh-v1111

Anima 2.9B Tiled High-Resolution Redraw

Purpose: take an existing image to a much larger size by upscaling it, tiling it, and running a low-denoise redraw per tile with the Anima 2.9B checkpoint. This is the "chunked" high-resolution route: the model never sees the whole canvas at once, so memory stays bounded while detail is rebuilt tile by tile.

Graph (as saved)

27 nodes / 35 links.

• UNETLoader Anima-2.9B-preview-v1.safetensors; CLIPLoader qwen_3_06b_base.safetensors (type stable_diffusion); VAELoaderKJ qwen_image_vae.safetensors.
• Input: LoadImage, then UpscaleModelLoader 4x_NMKD-Siax_200k.pth + ImageUpscaleWithModel, then ImageScaleToTotalPixels (lanczos, 4.0 MP).
• Tiling: TTP_Tile_image_size at 2x2 tiles with 0.35 overlap, TTP_Image_Tile_Batch / TTP_Image_Assy for batch and reassembly.
• Redraw: VAEEncode, KSampler 28 steps / CFG 5 / res_multistep / linear_quadratic / denoise 0.20, then VAEDecodeTiled.
• Caption helper: WD14Tagger|pysssss (wd-eva02-large-tagger-v3) + ShowText + string concat feed the prompt path.
• SaveImage prefix Anima_2.9B/tiled_redraw.

How to use

Start from a clean source image. The 0.20 denoise is deliberately low: it adds detail without repainting the composition, and raising it will make tiles disagree with each other. The tile overlap is what hides seams; if you change the tile count, keep the overlap and re-check the seams.

Limits and not-verified items

No inference was run for this listing: no speed, VRAM, seam-quality or resolution claim is made; every value is quoted from the saved nodes. Weights, upscale model, tagger, VAE, encoder and input images are not bundled. The cover image is a promotional still, not a frame produced by this graph.

Tutorial: YouTube · Bilibili. Workflows supplied through AI-KSK; verify workflow, node, model, upscaler, tagger, image and cover rights separately. No blanket commercial or redistribution license is asserted.