TenStrip/10Eros-Max

Shifted base H3, a capable model that still has base model capability and functions. Betas are all testing integrations and bases and are just samples ahead of inegrating sulphur h3 which will be the main data backbone and v1.0.

WORKFLOW: https://huggingface.co/LokkenJP/TenEros_Max_optimized_learned/tree/main/SampleWorkFlowTwoPassOptimized

12/3 shift, 8 step Euler/Simple or Er/SGM to start with on "8stepTurbo" versions. "raw" version is non-turbo for use with pika or other turbos.

beta_7 is a more correct format for average usability and combination of all working strategies so far. The merge is combination of the best lora data avaliable, premerged and then converted into MLP-only versions that were merged with magnitude cap, leaving full base model attention and reference intact. Isolating the merge to Gate-side data in MLP only allows for full base model behavior with extra data only needed to be reach with correct prompts. Should H3 loras all be trained only in MLP, and at much lower ranks? Probably if you ask me. The base is an fl2va model with the Fast Omni timestep, then refolded, allowing full hybridization and not the fake half-model kind of hybrid without a timestep. Does all modes with one key drawback being the need for VSA and 90% sparsity and some undiagnosed issues from the Omni structure; it is prone to 'flickering' in the background and on the latent in workflows with single pass and without refined sampling setups like 2-pass sampling. The 3D spatial upscale pass smooths the latent and removes it. The 8 step turbo covers 6-12 steps and is devoid of most turbo quality reductions aside from some more jagged noise, a tradeoff for letting v-band stay fully base to avoid contrast gains from turbo and preserve input quality, sampling on most normal setups, but for optimal quality Euler/Simple or Er_sde/SGM_uniform are reliable.

beta_6 is a rough test of a different merge strategy that shoud be avoided and either choose 5 or 7. It is built mostly-reference hybrid with a custom turbo meant to mimic how Fastv2 looks and sounds. Samples on 10/3 shift, fully dense or with Fastv2 VSA sparse node settings on multiple samplers like er_sde, euler, LCM, res_multistep, all having some different behaviors with prompt. It's a slow 8-12 step reference model in that configuration. 4step turbo loras can be run on top of it to reduce that. SLA sparsity seems to destroy it, avoid using that on any workflows. It's ideal configuration is with full gates and Fastv2 turbo and VSA gates when they have a reference solution figured out. The attempt to do that is the unstable version which has multiple issues but looks and sounds perfect at 8 steps.

beta_5 is built with a different normalization technique. It's built off 7 different concept-grouped grafts and consensus merges from over 20 loras, no direct lora merges. Lora credits are on the civit page. It allows for a non-turbo version which is avaliable. The files with TURBO have a hybrid turbo delta fusion baked in, saving 4.2 gigabytes of memory instead of having to load both ref/fl turbos. Other concept loras will also load on top more readily, usually needing lower strength 0.2-0.6.

Sampling and prompt are the key to quality and success. Non-turbo full step audio will always be better than audio on turbo versions.

Some working sampling setups:

er_sde/beta 4 steps

er_sde/beta57 4-6 steps

res_multistep/simple 6-8 steps (my go-to for best motion quality)

LCM/simple 6-8 steps (best turbo audio)

LCM/beta 4-6 steps

Euler/simple 4-8 steps

Int8 versions https://huggingface.co/cicalooo/10Eros-Max-h3-int8-convrot

beta_4 and beta_3 are corrupted test versions. beta_5 is the only true functional one so far. No cache or spectrum if doing reference, they cause accuracy loss. Hybrid model is built on the delta1024 h3 merge.

Regarding the grafts:

Due to a ton of confusion I've let Claude compile a full MD on the grafting, including code and methodology as it was mostly used to create code, document, and manage the graft project while I tested, architected, and mixed. I was keeping that to myself since maybe it'd actually be nice to have something to myself but after a handful of ignorant comments I'm open sourcing everything from it except the scripts themselves. You can hand the md back to claude or any competent agent and graft in the same way or have them explain it until you understand. No more dumbassery allowed now.

This is an evolving project subject to future fixes in H3 training. Training is clearly problematic, and my branch will rely on Sulphur H3 tunes. But I didn't want to wait for all that so I took the data from older models, grafting it in where the model needs it, grafting it to attn layers at a low level that doesn't disturb H3's visual or audio output quality. That's essentialy it.

The standard H3 community license applies. Because this release now carries transferred character from LTX 2.3, Wan 2.2, and Krea 2, the community licenses for those source models apply as well to the portions of character that came from each.