unify cross-links (both networks + both train repos)

embed speed chart + consumer note + cross-links

add HEART vs HAT_M speed benchmark (PyTorch/ONNX/TensorRT)

examples wording: full-res Real-ESRGAN set

link Real-ESRGAN N-vs-H comparisons

link NERVE-vs-HEART comparisons

document 1:1 crops + inputs

soften wording (recommended, not mandatory)

fix bf16 inference speed/VRAM claim to match measured benchmark

reorder sections for reading flow

enrich model-card metadata (library_name, tags, dataset)

mention ONNX export config in layout

rewrite Why section: academic tone, trainer motivation, community credit

move + expand 'Why HEART?' to second section

add structured 'help wanted' section

document what we tested / what we did not

tighten quickstart wording

no unverified claims, CC0 data section, embedded examples

simplify trainer quickstart (official converter)

document real-image example set

add 'What we actually tested' section

add BODY suite + relation to NERVE

Restructure: ADHD-friendly README, experiments doc, visual examples, official 4x/2x pretrains

Add final OTF v2 + GAN lq96 models and slimmed dynamic ONNX

HEART release package: arch, configs, ONNX, README, license