Community DiT
This is an experimental merged checkpoint based on Anima-Base v1.0.The model was created by merging Anima-Base with my own trained LoKR / LoRA weights, and also includes additional merged influence from hinaAnima2_R_Tweaker_v3.The inclusion of hinaAnima2_R_Tweaker_v3 has been permitted by the original R Tweaker creator.This is not an official Anima release, not a from-scratch base model, and not a standalone LoRA. It should be treated as an experimental derivative checkpoint based on Anima.Model overviewType: Checkpoint / merged checkpointBase model: Anima-Base v1.0Components:Anima-Base v1.0personally trained LoKR / LoRA weightshinaAnima2_R_Tweaker_v3 merged into the checkpoint with permission from the original R Tweaker creatorMain focus:anime illustrationmixed anime stylecharacter-oriented generationexperimental merge behaviorpossible partial 2.5D / depth-oriented rendering depending on prompt wordingTrigger word: No dedicated trigger word requiredYou can use descriptive prompts, Danbooru-style tags, or a mix of both.Because hinaAnima2_R_Tweaker_v3 is already merged into this checkpoint, loading it again is not required unless you intentionally want to amplify its effect.Version notesv005 unturboThis version explores a non-turbo / unturbo checkpoint direction based on Anima-Base.The goal is to preserve the flexibility of Anima while adding my own trained style and character tendencies through LoKR / LoRA merging. The final checkpoint also includes additional R Tweaker v3 influence, which may help push the output toward a more dimensional anime style depending on the prompt.Compared with a normal LoRA workflow, this checkpoint is intended to provide a more direct merged output without requiring a separate LoRA weight setting.Expected behavior:anime-style illustration outputflexible prompt responsestronger personal mixed-style tendency than the original baseusable with both tag-based and natural language promptsmay show partial 2.5D / semi-realistic anime behavior depending on prompt wordingRecommended settings without Turbo LoRAUse these settings if you want to run the checkpoint directly without Anima Turbo LoRA.Sampler: er_sde / euler_a / dpmpp_2m_sde_gpu / Res MultistepScheduler: Simple or SGM_UniformSteps: 25–50CFG Scale: 4–5Resolution: 512x512 to 1536x1536For quick testing, 25–30 steps can work. For final quality evaluation, 30–50 steps is usually safer.Recommended settings with Anima Turbo LoRAThis checkpoint can also be used together with Anima Turbo LoRA for faster generation.Recommended Turbo LoRA:https://civitai.red/models/2560840/anima-turbo-loraWhen using Anima Turbo LoRA, you should lower both CFG and step count.Suggested Turbo settings:Anima Turbo LoRA weight: 0.7–1.0Steps: 8–12CFG Scale: around 1Sampler: Euler is a safe starting pointScheduler: Simple or SGM_UniformDo not use normal high-CFG / high-step settings together with Turbo LoRA at first. Turbo LoRA is designed for short-step and low-CFG generation.If the image becomes too overcooked, too flat, too noisy, or too unstable, reduce the Turbo LoRA weight, lower CFG, reduce steps, or simplify the prompt.Prompt tipsStart with quality and safety tags:masterpiece, best quality, score_7, safe, highres For negative prompt, a short negative prompt is usually enough:worst quality, low quality, score_1, score_2, score_3, artist name In many cases, adding too many negative tags may reduce variety or over-correct the image. Start simple first, then add more negative tags only when a specific issue appears.Example prompt stylemasterpiece, best quality, score_7, safe, highres, 1girl, anime illustration, detailed eyes, soft lighting, looking at viewer, upper body, clean background For more 2.5D / depth-oriented output, you can try adding terms such as:cinematic lighting, soft shadows, depth of field, detailed shading, semi-realistic anime style For a flatter anime-style output, avoid too many realistic rendering terms and use simpler anime illustration tags.Notes and limitationsThis is an experimental merge model. Output may vary depending on prompt length, seed, sampler, scheduler, resolution, CFG, and any additional LoRAs used.Known limitations:exact character outfit accuracy is not guaranteedhands and small details may still require prompt refinement or inpaintingshort prompts may produce unstable or overly generic resultstext rendering is not a main strengthadditional LoRAs may conflict with the merged styleusing R Tweaker v3 again on top of this checkpoint may over-amplify its effectusing Turbo LoRA requires lower CFG and fewer steps than normal Anima settingsCreditsBase model:Anima-Base v1.0 by CircleStone Labs / Comfy OrgMerged influence:my own trained LoKR / LoRA weightshinaAnima2_R_Tweaker_v3, used with permission from the original R Tweaker creatorRecommended optional acceleration LoRA:Anima Turbo LoRAhttps://civitai.red/models/2560840/anima-turbo-loraLicense and usageThis model is based on Anima-Base v1.0 and should follow the original Anima license restrictions.Non-commercial use only.This checkpoint includes merged influence from hinaAnima2_R_Tweaker_v3, used with permission from the original R Tweaker creator.Please note that this is an independent experimental merge. It is not an official Anima release, not an official R Tweaker release, and not an official release from CircleStone Labs, Comfy Org, NVIDIA, or any original IP holder.This model is not affiliated with, endorsed by, sponsored by, or officially connected to CircleStone Labs, Comfy Org, NVIDIA, or any original IP holder.