Anima Delta Mix — 2B into 2.9B / 3.8BWrite into official 2.9B / 3.8B insert slots without souping the stem.Node pack: <a target="_blank" rel="ugc" href="https://github.com/bestluner-create/Anima-Delta-Mix">https://github.com/bestluner-create/Anima-Delta-Mix</a> Stake dump (current zip): <a target="_blank" rel="ugc" href="https://github.com/bestluner-create/Anima-Delta-MixV2">https://github.com/bestluner-create/Anima-Delta-MixV2</a>This is not Remap and not a normal merge. Remap answers which address a tensor should land on when you move 28→40→52. Delta Mix answers what may be written at that address.The floor map is shared (official 2.9 inserts and the 3.8 trainable_dit_blocks). The write rule is not.IMPORTANT<ul><li>Donors = 28-block 2B only. Never drop 2.9 / 3.8 into donor_stem / donor_gen1 / donor_gen2.</li><li>Two chassis, two nodes. Do not mix them.<ul><li>0.6B line: qwen_3_06b_base. Node DeltaMix-3.8B 52.</li><li>Q35 line: Anima 3.8B v1.1 + Connector. Node DeltaMix-3.8B-Q35. Runtime Attach/Prompt live in a separate pack.</li></ul></li><li>Bake with ModelSave, not CheckpointSave.</li><li>Compatible implementation. Not unlock / bypass / “better than official 1.1”.</li></ul>What we refuse<ul><li>No average of two checkpoints inside one block.</li><li>No smearing a donor across the wrong indices.</li><li>The shelf is not “the last 12 blocks”. The stem is the 28 inherited floors. They sit at the start of the tape and in the tail. What we do</li></ul><pre><code>stem 28 inherited slots freeze (w=0) or lerp
insert official expansion insert += r · (donor[src] − kept)
gate unique skip drop the whole floor if donor ≈ stem</code></pre>src is the mapped 2B index. An empty niche is better than the wrong floor. Nodes (menu: Anima Delta Mix)<pre><code>DeltaMix-2.9B 40 chassis 40 · 0.6B stem + Gen-1
DeltaMix-3.8B 52 chassis 52 · 0.6B stem + Gen-1 + Gen-2
DeltaMix-3.8B-Q35 chassis 52 · v1.1 stem + Gen-1; G2 factory OFF
Lora3.8B_only52 52 MODEL only 28/40/52 LoRA onto the stem
 skip_gen2 ON</code></pre>Class ids are unchanged so old graphs still load.On 3.8 the two shelves have different jobs: Gen-1 = crop, Gen-2 = trained surface (not the same merge). G2 unique is checked vs stem and vs G1. Do not put one branch on both shelves. <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/35eb924b-bee5-4ff2-ae2c-2cf85ba4473b/original=true/35eb924b-bee5-4ff2-ae2c-2cf85ba4473b.jpeg" /> <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0d668235-78f5-4551-9661-3fa66a46f98d/original=true/0d668235-78f5-4551-9661-3fa66a46f98d.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1e43802e-8e31-40c8-a363-62de1f4e96c1/original=true/1e43802e-8e31-40c8-a363-62de1f4e96c1.jpeg" /> <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3aa266ba-2b65-4ea6-84d9-598de1ecc858/original=true/3aa266ba-2b65-4ea6-84d9-598de1ecc858.jpeg" /> <img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/78f2cc5e-1ab1-4f96-b323-c4a90a893263/original=true/78f2cc5e-1ab1-4f96-b323-c4a90a893263.jpeg" /> Q35 extras<ul><li>Connector + native llm_adapter stay on the chassis.</li><li>Watch KEEP [7, 15, 23, 31].</li><li>No copy_source on this node.</li><li>Optional G2 test only: inject_delta ~0.35 on writable slots. Not the factory default.</li></ul>How to run it<ul><li>Load 2.9 or 3.8 into the matching node.</li><li>Point file fields at 2B donors on disk (low-RAM path).</li><li>Unique ON. Neighbor / spare / deep OFF unless you are experimenting.</li><li>w=0 leaves the official stem. w&gt;0 is already a different checkpoint, not “a bit more quality”.</li><li>12GB: queue one group at a time.</li><li>52-file that loads as 28 blocks: install comfyui-anima-3-8B / use portable. Console before the graph.</li><li>Preview latent: 16ch (EmptySD3LatentImage). Shift 3–4.5. A starting point, not dogma</li></ul><pre><code>2.9 / 40 G1 inject_delta 0.35 unique 0.05
3.8 0.6B G1 inject_delta 0.35
 G2 reset_muted_then_delta 0.45
 unique 0.05
Q35 W=0 G1 inject_delta 0.35 G2 off</code></pre>Styled 0.6B V3 used copy_body_keep_adaln on both shelves (adaLN stays chassis). That is a bake choice, not the factory default of the node. How to read the log<pre><code>dropped=['14:vs_stem'] duplicate — slot not written
holes=6 empty niches — expected
G2 vs_D1 on all 12 same crop as G1 — change the donor, do not crank r</code></pre>dropped is the method, not a crash.The point 2.9 already owns ~12 insert floors. 3.8 owns another 12. The node does not mint new billions of weights. It occupies houses that already exist, using a 2B donor, and throws away a floor that adds nothing. A 28-LoRA after the mix still lands where it would on base. Why this is not a merge<pre><code>full-DiT merge / lerp every block is soup of two files
extend_to_new_layers LoRA smeared onto inserts as well
Delta Mix soup banned in the niche
 niche gets a delta to kept
 a duplicate is dropped whole</code></pre>CreditsAnima — CircleStone Labs.2.9 / 3.8 maps and 3.8B v1.1 Connector — lylogummy.Write policy — Delta Mix.Anima license is non-commercial.

Files with SHA256: af59f01f57bb10cd4054748002472dc3924acfec0292dd8512aa4ebf2990cd46