{"rewrite":{"id":"r_951c2a260981442c8600747f","clusterId":"c_2d8b6bd1eff64e3eafc3829a","slug":"anima-base-v1-0-image-generation-ai-model-released-for-local-anime-art","model":"deepseek-v4-flash","headline":"Anima Base v1.0 Image Generation AI Model Released for Local Anime Art","summary":"On May 15, the image generation AI model Anima Base v1.0, co-developed by ComfyUI, was released. With 2 billion parameters and a 4.18GB file size, it runs on local PCs with 8GB VRAM and supports resolutions up to 1536x1536. The model is specialized for anime style and aims to improve consistency for character generation, a known weakness in previous open models.","whyItMatters":"The release marks a step toward practical anime-style image generation on local hardware, addressing the consistency problem that has limited open models for character art.","webCardHtml":"\u003cp\u003eAnima Base v1.0, co-developed by ComfyUI, launched on May 15 as a local image generation model specialized for anime. The model has 2 billion parameters and a 4.18GB file size, small enough to run on video cards with 8GB VRAM. It supports resolutions up to 1536x1536, larger than the common 1024x1024. On an A6000 GPU, generating a single 1024x1024 image takes about 30 seconds.\u003c/p\u003e\u003cp\u003eThe model was released as a preview version in January by US-based Circle Stone Lab, which receives funding from ComfyUI, and has been updated monthly. The v1.0 base completed model has reduced noise compared to the preview version, allowing clearer image output. However, the art style still fluctuates significantly with each generation using the same prompt, making consistency difficult to control with prompts alone.\u003c/p\u003e\u003cp\u003eTo address this, LoRA training is recommended. The sd-scripts tool by Japanese engineer Kohya Tech supported Anima in February, and the Anima Standalone Trainer, a GUI version by gazingstars, is now available for easier LoRA creation.\u003c/p\u003e","blueskyPost":"Anima Base v1.0's ability to run on 8GB VRAM while supporting 1536x1536 resolution is unusual for a specialized anime model, but its art style fluctuation per generation makes LoRA training essential.","twitterPost":"Anima Base v1.0 runs on 8GB VRAM at 1536x1536, but its style fluctuation per generation makes LoRA training necessary.","threadsPost":"Anima Base v1.0 is a 2B-parameter anime model that runs on 8GB VRAM and supports 1536x1536 resolution. Its art style fluctuates significantly with each generation, so LoRA training is key to maintaining character consistency.","newsletterBlurb":"Anima Base v1.0, an anime-specialized image generation AI co-developed by ComfyUI, launched on May 15. The 2B-parameter model runs on local PCs with 8GB VRAM and supports resolutions up to 1536x1536. While the base model reduces noise compared to preview versions, art style consistency remains a challenge, addressed through LoRA training tools now available.","attributionJson":"[{\"source\":\"ASCII.jp\",\"url\":\"https://ascii.jp/elem/000/004/404/4404597/?rss\",\"title\":\"What's Next After SDXL? The Surprising Power of a Local Image Generation AI Specializing in Anime (1/5)\"}]","lintFlagsJson":"[]","lintHits":0,"costUsd":0,"inputTokens":4192,"outputTokens":758,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1779826512,"createdAt":"2026-05-25T00:02:41.000Z","publishedAt":"2026-05-25T00:16:01.000Z","updatedAt":"2026-05-25T00:16:01.000Z"},"cluster":{"id":"c_2d8b6bd1eff64e3eafc3829a","canonicalTitle":"SDXLの次はこれ？　アニメ特化のローカル画像生成AI、驚きの実力 (1/5)","representativeArticleId":"a_991e66c997e3142a15f89917","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-05-24T22:00:00.000Z","lastSeenAt":"2026-05-24T22:00:00.000Z","updatedAt":"2026-05-25T00:16:01.000Z"},"attribution":[{"source":"ASCII.jp","url":"https://ascii.jp/elem/000/004/404/4404597/?rss","title":"SDXLの次はこれ？　アニメ特化のローカル画像生成AI、驚きの実力 (1/5)"}],"entities":{"anime_titles":[],"manga_titles":[],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["Anima Base v1.0 launched on May 15, co-developed by ComfyUI, with 2 billion parameters and a 4.18GB file size.","The model runs on local PCs with 8GB VRAM and supports resolutions up to 1536x1536.","On an A6000 GPU, generating a single 1024x1024 image takes about 30 seconds.","The v1.0 base model has reduced noise compared to the preview version, but art style still fluctuates significantly with the same prompt.","LoRA training is recommended for consistency; the Anima Standalone Trainer by gazingstars is now available for easier LoRA creation."]}