<h1 id="comfyui-anima-legacy-universal-lora-loader">ComfyUI Anima Legacy / Universal LoRA Loader</h1><p>A ComfyUI compatibility loader for using LoRAs across different generations of the <strong>Anima DiT architecture</strong>.</p><p>This project does <strong>not</strong> introduce or claim authorship of <strong>Anima 3.8B</strong>, the <strong>52-block expansion</strong>, the <strong>Qwen 0.6B + Qwen 3.5 dual-encoder approach</strong>, <strong>expanded conditioning</strong>, or <strong>Semantic Connector / Semantic Connector v2</strong>.</p><p>Those architectures and techniques originate from the original <strong>Anima 3.8B / Anima Expanded 3.8B</strong> work.</p><h2 id="credits-attribution">Credits / Attribution</h2><p><strong>Anima Expanded 3.8B</strong><br /><a target="_blank" rel="ugc" href="https://civarchive.com/models/2910974/anima-expanded-38b">https://civarchive.com/models/2910974/anima-expanded-38b</a></p><p><strong>Anima 3.8B</strong><br /><a target="_blank" rel="ugc" href="https://civarchive.com/models/2880480/anima-38b">https://civarchive.com/models/2880480/anima-38b</a></p><p><strong>Original Hugging Face repository</strong><br /><a target="_blank" rel="ugc" href="https://huggingface.co/lylogummy/Anima-3.8B">https://huggingface.co/lylogummy/Anima-3.8B</a></p><p>The contribution of this custom node is limited to <strong>LoRA compatibility and runtime block remapping across different Anima generations</strong>.</p><p>The original version was created to use <strong>legacy 28-layer Anima LoRAs</strong> with <strong>40-layer Anima 2.9B / expanded Anima checkpoints</strong>.</p><p>Starting with <strong>v1.4</strong>, the loader adds compatibility for <strong>Anima 3.8B / 52-layer checkpoints</strong> and can automatically handle <strong>28-layer, 40-layer, and 52-layer Anima LoRAs</strong>.</p><p>No LoRA file is permanently converted or modified. The compatibility remapping is performed at runtime when the LoRA is loaded in ComfyUI.</p><hr /><h1 id="versions">Versions</h1><h2 id="v1.4-universal-28l-40l-52l-loader">v1.4 — Universal 28L / 40L / 52L Loader</h2><p>Recommended for <strong>Anima 3.8B / 52-layer checkpoints</strong>.</p><p><strong>New node:</strong> <code>Anima38AutoLoRALoader</code></p><p><strong>Display name:</strong> <code>Anima 3.8B Universal LoRA Loader (Auto 28L / 40L / 52L)</code></p><p>The loader automatically selects the required conversion path:</p><p><strong>Legacy Anima 28L LoRA → 28L → 40L → 52L → Anima 3.8B</strong></p><p><strong>Anima 2.9B 40L LoRA → 40L → 52L → Anima 3.8B</strong></p><p><strong>Anima 3.8B 52L LoRA → Direct load → Anima 3.8B</strong></p><p>The existing legacy loader node IDs are retained for workflow compatibility.</p><h2 id="v1.0_28l_to_40l">v1.0_28L_to_40L</h2><p>Original compatibility version for:</p><p><strong>Legacy Anima 28L LoRA → 28L → 40L → Anima 2.9B</strong></p><p>Use this version if you only need legacy 28-layer LoRAs with <strong>Anima 2.9B / 40-layer checkpoints</strong>.</p><hr /><h1 id="v1.4-main-features">v1.4 Main Features</h1><ul><li><p>Automatic source LoRA detection: <strong>28L / 40L / 52L</strong></p></li><li><p>Legacy <strong>28L → 40L → 52L</strong> compatibility</p></li><li><p>Native Anima 2.9B <strong>40L → 52L</strong> compatibility</p></li><li><p>Native Anima 3.8B <strong>52L direct loading</strong></p></li><li><p>Optional weak LoRA propagation into newly inserted blocks</p></li><li><p>Sharp / anti-blur oriented propagation modes</p></li><li><p>Manual source architecture override</p></li><li><p>Detection report with architecture, confidence, and conversion path</p></li><li><p>Compatibility with older workflow JSON files</p></li><li><p>No permanent modification of the original LoRA</p></li><li><p>No checkpoint or LoRA weights included</p></li></ul><hr /><h1 id="automatic-lora-type-detection">Automatic LoRA Type Detection</h1><p>With <code>source_type = auto</code>, the v1.4 loader examines the LoRA structure before loading it.</p><h2 id="basic-detection-rules">Basic Detection Rules</h2><p><strong>Highest block index 40–51 → Anima 3.8B / 52L LoRA</strong></p><p><strong>Highest block index 28–39 → Anima 2.9B / 40L LoRA</strong></p><p><strong>Only block indices 0–27 → Legacy 28L LoRA candidate</strong></p><p>If the tensor structure alone is ambiguous, the loader also checks available <strong>safetensors metadata</strong> and filename hints.</p><p>The report shows:</p><p><code>Detected architecture / Detection confidence / Detected block indices / Selected conversion path / Number of converted LoRA entries</code></p><h2 id="important-note-about-partial-loras">Important Note About Partial LoRAs</h2><p>Some LoRAs only train a limited number of blocks.</p><p>For example, a real Anima 3.8B LoRA may contain only blocks <code>0–15</code>. In that case, its block structure alone can look identical to a legacy 28-layer LoRA.</p><p>Because this cannot always be determined perfectly from tensor keys alone, v1.4 also provides a manual override:</p><p><code>source_type: auto / 28L / 40L / 52L</code></p><p>Normally, use <code>auto</code>.</p><p>If the report shows a low-confidence detection for a partial LoRA, manually select the correct architecture.</p><hr /><h1 id="28-40-block-mapping">28 → 40 Block Mapping</h1><p><code>0→0, 1→1, 2→3, 3→4, 4→6, 5→7, 6→9, 7→10, 8→12, 9→13, 10→15, 11→16, 12→18, 13→19, 14→20, 15→22, 16→23, 17→25, 18→26, 19→28, 20→29, 21→31, 22→32, 23→34, 24→35, 25→37, 26→38, 27→39</code></p><h2 id="inserted-anima-2.9b-blocks">Inserted Anima 2.9B Blocks</h2><p><code>2, 5, 8, 11, 14, 17, 21, 24, 27, 30, 33, 36</code></p><hr /><h1 id="40-52-expansion">40 → 52 Expansion</h1><p>For Anima 3.8B, twelve additional blocks are inserted into the 40-layer layout.</p><h2 id="inserted-anima-3.8b-blocks">Inserted Anima 3.8B Blocks</h2><p><code>3, 7, 11, 15, 19, 23, 27, 31, 35, 39, 43, 47</code></p><p>The original 40 Anima 2.9B blocks are placed into the remaining positions in order.</p><p>Therefore a legacy LoRA follows:</p><p><strong>28L → 40L → 52L</strong></p><p>instead of attempting a direct 28 → 52 conversion.</p><hr /><h1 id="recommended-starting-settings">Recommended Starting Settings</h1><h2 id="legacy-28l-lora-anima-3.8b-52l">Legacy 28L LoRA → Anima 3.8B 52L</h2><p><strong>Model strength:</strong> <code>0.75</code></p><p><strong>28 → 40:</strong> <code>copy_prev / 0.18 / attn_only</code></p><p><strong>40 → 52:</strong> <code>copy_prev / 0.12 / attn_only</code></p><h2 id="native-anima-2.9b-40l-lora">Native Anima 2.9B / 40L LoRA</h2><p><strong>Model strength:</strong> <code>0.75</code></p><p><strong>40 → 52:</strong> <code>copy_prev / 0.08–0.16 / attn_only</code></p><h2 id="native-anima-3.8b-52l-lora">Native Anima 3.8B / 52L LoRA</h2><p><strong>Source type:</strong> <code>auto</code></p><p>No block expansion is performed. The LoRA is loaded directly at the selected <code>strength_model</code>.</p><hr /><h1 id="sharp-anti-blur-modes">Sharp / Anti-Blur Modes</h1><h2 id="copy_prev">copy_prev</h2><p>Copies the LoRA contribution from the previous source block.</p><p><strong>Recommended starting mode for sharper results.</strong></p><h2 id="sharp_blend">sharp_blend</h2><p>Approximately <strong>70% previous block / 30% next block</strong>.</p><h2 id="blend">blend</h2><p>Approximately <strong>50% previous block / 50% next block</strong>.</p><p>This may produce smoother transitions, but can also make the result softer.</p><h2 id="copy_next">copy_next</h2><p>Uses the next source block.</p><h2 id="disabled">disabled</h2><p>No synthetic LoRA contribution is applied to inserted blocks.</p><hr /><h1 id="propagation-scope">Propagation Scope</h1><h2 id="attn_only">attn_only</h2><p>Only attention-related LoRA components are propagated.</p><p><strong>Recommended for the first test.</strong></p><h2 id="attn_and_mlp">attn_and_mlp</h2><p>Propagates both attention and MLP-related components.</p><h2 id="all">all</h2><p>Propagates all supported LoRA components.</p><hr /><h1 id="workflow-compatibility">Workflow Compatibility</h1><p>The following older node IDs remain available:</p><p><code>Anima29LegacyLoRALoader</code></p><p><code>Anima29LegacyLoRALoaderSharp</code></p><p><code>Anima38LegacyLoRALoader</code></p><p><code>Anima38LegacyLoRALoaderSharp</code></p><h2 id="v1.4-universal-node">v1.4 Universal Node</h2><p><code>Anima38AutoLoRALoader</code></p><h2 id="compatibility-alias">Compatibility Alias</h2><p><code>Anima38UniversalLoRALoader</code></p><p>This allows older workflow JSON files to continue loading while new Anima 3.8B workflows can use automatic LoRA architecture detection.</p><hr /><h1 id="what-this-project-does-and-does-not-do">What This Project Does — and Does Not Do</h1><p>This custom node provides <strong>runtime LoRA compatibility and block remapping</strong>.</p><p>It does <strong>not</strong> create or replace the underlying Anima architecture.</p><p>In particular, this project does <strong>not</strong> claim authorship of:</p><ul><li><p>Anima 3.8B</p></li><li><p>the Anima 3.8B 52-block architecture</p></li><li><p>the 40-block → 52-block Anima expansion concept</p></li><li><p>the Qwen 0.6B + Qwen 3.5 dual-encoder approach</p></li><li><p>expanded conditioning</p></li><li><p>Semantic Connector</p></li><li><p>Semantic Connector v2</p></li><li><p>the underlying Anima 3.8B model or text-conditioning architecture</p></li></ul><p>For those developments, please refer to the original projects listed in the <strong>Credits / Attribution</strong> section above.</p><p>The purpose of this custom node is specifically to make LoRAs trained for different Anima block layouts more convenient to use with expanded Anima checkpoints.</p><hr /><h1 id="installation">Installation</h1><p>Extract the package so that the node is located here:</p><p><code>ComfyUI/custom_nodes/ComfyUI-Anima29-LegacyLoRA/__init__.py</code></p><p>If an older version is already installed, replace the existing <code>ComfyUI-Anima29-LegacyLoRA</code> folder.</p><p>Then fully restart ComfyUI.</p><hr /><h1 id="important-notes">Important Notes</h1><p>This custom node does <strong>not</strong> retrain the LoRA.</p><p>It performs compatibility remapping only when the LoRA is loaded into ComfyUI.</p><p>Results may differ from the original source architecture because newly added Anima blocks did not exist when older LoRAs were trained.</p><p>Compatibility also depends on how the LoRA was trained. Unusual LoRA key formats or highly selective partial-block LoRAs may require manual <code>source_type</code> selection.</p><p>Automatic detection is intended as a convenience feature, not a guarantee that every unusual or partial LoRA can be identified correctly.</p><p>This is an experimental compatibility tool intended to make LoRAs from multiple Anima generations easier to use with expanded Anima checkpoints.</p><hr /><h1 id="credits">Credits</h1><p>The Anima 3.8B architecture, expanded architecture work, dual-encoder conditioning approach, and related conditioning techniques are credited to their original developers.</p><p><strong>Anima Expanded 3.8B</strong><br /><a target="_blank" rel="ugc" href="https://civarchive.com/models/2910974/anima-expanded-38b">https://civarchive.com/models/2910974/anima-expanded-38b</a></p><p><strong>Anima 3.8B</strong><br /><a target="_blank" rel="ugc" href="https://civarchive.com/models/2880480/anima-38b">https://civarchive.com/models/2880480/anima-38b</a></p><p><strong>Original Hugging Face repository</strong><br /><a target="_blank" rel="ugc" href="https://huggingface.co/lylogummy/Anima-3.8B">https://huggingface.co/lylogummy/Anima-3.8B</a></p><p>This custom node only provides the <strong>LoRA compatibility / runtime remapping implementation</strong> described above.</p><hr /><h1 id="license">License</h1><p><strong>MIT License</strong> for the custom node code.</p><p>No checkpoint weights, LoRA weights, or third-party model files are included.</p>

Files with SHA256: b24fe3bfa9473cc2895f39d2a910e24d42e9a4d4b9f79679eddd681a5e3aff0e