Hello there, finally attacking Z image base full finetune. it is complexe but interesting. This model add nudity, anatomy, and variations into the model, helping with proportions posing and angle, adding concepts like genitalias, and also give some push in inbetween body shapes and expressions, ranging from chubby, slender, soft, mucular, realistic proportions, to extreme proportions. This model, can make now, nude men, nude feminine men, nude transwomen, and nude womens. call them like that and experiment, there is not real trigger words, the captioning was donne, with Kimi K2.5 for the natural language part, and I added a tag line with Joytag, tout have some repeating concepts to help the texte encoder to grab some recurrent elements. so prompt in a mix of natural and tags. Should be able to do a lot of various styles. Shift 5, CFG 4, eular Beta, and 40-50 steps for seeing good results. It work even better with the lora distil fun 4 steps. with CFG 1, 4 or 6 steps Still imperfect obviously, V2 is in the back of my mind, with larger Dataset, some better selections in mind, and a deeper, more hard funel technics and some test about BF16 precisions. still adjusting. Always learning(Human wrote part)<h1 id="z-image-base-full-finetune-technical-notes-(experimental-observations)"> Z-Image Base Full Finetune – Technical Notes (Experimental Observations)</h1>This model was trained as a full finetune on Z-Image Base (BF16) using Musubi Tuner.Training was performed on an H100 80GB, at 1024 resolution, with:Batch size: 9Gradient accumulation: 1Full BF16Flash attention enabledGradient checkpointingBucketed dataset (many varied buckets)Dataset of 2100 Images, with very various styles, poses, angles and variations in physiologies.The following notes are based on practical experimentation.They are not official documentation, and should be considered empirical observations that worked in this setup.<h2 id="1.-z-image-base-behaves-differently-from-sd-models">1. Z-Image Base behaves differently from SD models</h2>Z-Image Base uses a DiT transformer architecture and flow-based timestep sampling.Compared to SD1.5 / SDXL:Structural changes take longer to appear.The model shows strong internal coherence.It resists abrupt shifts.Visible improvement often depends heavily on sampling settings.It feels layered — structural changes must propagate across multiple refinement stages.Z-Image Base is harder to move, but very stable once shaped.<h2 id="2.-about-shift-(flow-shift)-during-training">2. About shift (Flow Shift) During Training</h2>Using:--timestep_sampling shift--discrete_flow_shift Xmodifies how timesteps are distributed during training.From experimentation:Higher shift (≈ 2.5+)Emphasizes global structure.Useful for early structural imprinting.Mid shift (~2.0–2.2)Appears to consolidate structure.Balances geometry and detail.Lower shift (1.5–1.7)Seems to refine fine details.Useful for finishing phases.This suggests a staged approach:Start higher for structure, progressively lower for refinement.This is an experimental strategy — not an official rule.<h2 id="3.-optimizer-choice-(adafactor-vs-adamw)">3. Optimizer Choice (Adafactor vs AdamW)</h2>In this setup, Adafactor performed better than AdamW for full finetuning.Observed behavior:Lower VRAM usageLarger stable batch size (9 at 1024 on H100)More stable long-phase convergenceExample configuration used:--optimizer_type adafactor--optimizer_args "relative_step=False" "scale_parameter=False" "warmup_init=False"--lr_scheduler constant_with_warmupAgain, this is empirical — other setups may vary.<h2 id="4.-learning-rate-funnel-strategy">4. Learning Rate Funnel Strategy</h2>A progressive reduction strategy was used:Early phase: higher LR for structural changeMid phase: moderate LR for stabilizationFinal phases: very low LR for micro-refinementThe core idea:Move the structure first.Refine later without breaking global coherence.Z-Image Base appears to benefit from staged training rather than a flat learning rate schedule.<h2 id="5.-dataset-and-bucketing">5. Dataset and Bucketing</h2>The dataset was:1024 resolution aligned with Z-Image BaseFully bucketedWith many varied aspect bucketsMulti-distribution (varied morphologies)Using many buckets helped:Preserve structural consistencyAvoid overfitting to a narrow framingMaintain prompt flexibility<h2 id="6.-sampling-matters-more-than-expected">6. Sampling Matters More Than Expected</h2>Z-Image Base can look soft or “vaporous” under weak guidance.Under stronger guidance (e.g. 4.0) and sufficient steps (e.g. 50):Structural improvements become significantly clearer.Anatomical refinement is more visible.Prompt conditioning becomes stronger.When evaluating finetunes:Use consistent seedsTest guidance 3–5Try multiple flow_shift sampling values (e.g. 3 and 5)<h3 id="compare-phases-side-by-side-and-loss-is-not-that-telling-just-watch-it-did-not-spike-but-you-will-not-have-deep-dives-down-your-best-bet-is-sampling-between-phases-of-training.">Compare phases side-by-side, and Loss is not that telling, just watch it did not spike, but you will not have deep dives down, your best bet is sampling between phases of training.</h3>The model’s internal changes may not appear under weak sampling settings.<h2 id="7.-important:-these-are-hypotheses">7. Important: These Are Hypotheses</h2>Everything above is based on hands-on experimentation with:Musubi TunerZ-Image Base (BF16)H100 80GB1024 resolutionLarge batch (9)Progressive shift funnelZ-Image Base is complex.Different datasets, hardware, or goals may respond differently.These notes should be treated as:Practical observationsNot universal truthA starting point for experimentationThis finetune aimed to preserve:Z-Image Base’s native photorealistic grainStructural coherencePrompt responsivenessStability under guidanceIf you experiment further with Z-Image Base,structured training and careful sampling evaluation seem essential.(This note has been made by Ai, to avoid confusion, I am not a nativ speaker, but I reread it and approves and take the responsability of this experience return)

Files with SHA256: c3bda91f722b041f85ee53a2d5c3bcfe1cc14d7716b2dca0497e8068421a5545