TheMindExpansionNetwork/sonicforagecrew-ideogram-character-test-20

sonicforagecrew — character reference test

Synthetic concept art produced with the supplied custom Ideogram 4 workflow. This is a bounded test of 20 original identities: ten funny briefs and ten serious briefs. It is not a 3D asset pack, trained model, animation rig, or proof of general model/LoRA quality.

Files

• pairs/: unchanged generated PNG masters, same-stem .txt captions grounded in inspected pixels, and same-stem .json generation prompts.
• prompts/: structured character briefs and prompt inventory. Prompts describe what was requested, not necessarily what appeared.
• workflows/: byte-preserved original plus 20 API-format job workflows. Only prompt, seed and output prefix vary.
• manifest.jsonl: identity, category, prompt, exact filenames, seed, workflow hashes, accepted cloud job ID, decoded dimensions/image hash, and quality review.
• qa/: labeled contact sheet, grouped review images and explicit per-image notes. Rejected-for-brief results remain included as test evidence, not silently passed as clean training data.
• PREFLIGHT.md: node/model checks, preserved settings, quote caveats and execution controls.

What this test means

The shared trigger is exactly sonicforagecrew. It invokes an already loaded LoRA/style association. A trigger is not a unique character identity and does not guarantee the intended species, silhouette, layout or anatomy. Names here are original conceptual metadata, not independently trained identity tokens. Generating these images does not train, fine-tune or validate any model.

Caption pairs are a convenient dataset format, not a claim of training readiness. Review qa/visual_notes.json and exclude failed/malformed samples before any training. In particular, source-prompt lore is not used as a substitute for observed image details. No training or 3D conversion was performed. This repository publishes the generated test dataset; it is not a trained-model release.

Customize a character

1. Start with an item in prompts/ and keep prompt: sonicforagecrew unchanged for this family.
2. Change species and silhouette first: body mass, height, number and placement of limbs, head/body ratio, tail/wing geometry. Avoid making only palette variants of one shape.
3. Set costume construction (not only costume color), a small deliberate palette, and one or two props that reveal the role.
4. Put unchanging features in identity_constraints: proportions, limb count, markings, costume seams and accessories must agree across all views. A front-view shoulder patch must occupy the corresponding side in the back view.
5. Keep the layout contract explicit: front, side, back and three-quarter full-body views, plus separate complete action pose and prop studies; neutral backdrop, no generated writing, no crop.
6. Serialize the complete object with json.dumps into workflow node 199.inputs.text; do not paste malformed JSON-like fragments. Use a unique seed at 197.inputs.noise_seed and prefix at 187.inputs.filename_prefix. Keep the original workflow copy untouched.
7. Inspect the resulting pixels before writing a caption. A caption starts with sonicforagecrew, followed by the metadata name and visibly supported details. If the image does not match the intended species or clothing, caption what exists and mark the failure.
8. New generations or paid retries require their own approved bounded scope. Recover existing job IDs before any resubmission.

Preserved generation configuration

Quality 48 steps, Euler sampler, mu 0.0, std 1.5; DualModelGuider CFG 7; CFGOverride 3 over 0.7–1.0. LoRA applied to both diffusion branches at strength 1. The original 16:9 / 1MP selector and multiple-16 alignment are retained, so actual output dimensions may only approximate exact 16:9.

Exact model files:

• ideogram4_int8_convrot.safetensors
• ideogram4_unconditional_int8_convrot.safetensors
• qwen3vl_8b_fp8_scaled.safetensors (CLIP type ideogram4)
• flux2-vae.safetensors
• TheMindExpansionNetwork__sonicforagecrew_idogram_v1__sonicforagecrew_idogram_v1.safetensors (both branches)

These are the supplied filenames, including idogram spelling. No model/LoRA substitutions or downloads were made.

Costs and rights

The MCP quote is 0 paid API-node credits and explicitly excludes GPU/queue time and storage. GPU billing is unknown here: zero API-node credits does not mean zero compute cost. Exactly 20 accepted generations, no paid retries. The MCP batch call ran without a fresh confirmation gate; confirm:true was never supplied.

Original character briefs intentionally avoid existing franchise/IP names. Synthetic origin does not itself establish unrestricted commercial or model-training rights: review the base model, LoRA and provider terms for your intended use. No blanket rights warranty is made.

Video

The separate hyperframes-source.zip contains all required project media and composition source. The rendered slideshow displays every sheet complete with names outside the generated images, and no generated music. Generated defects remain visible; video framing does not repair them.

Download from Hugging Face

hf download TheMindExpansionNetwork/sonicforagecrew-ideogram-character-test-20 --repo-type dataset --local-dir sonicforagecrew-character-test

git clone https://huggingface.co/datasets/TheMindExpansionNetwork/sonicforagecrew-ideogram-character-test-20

Large files may require Git LFS; the hf download command fetches actual files. The pairs/ folder contains exactly 20 PNG/TXT/JSON triples. checksums.json records the original package; release-sha256.json covers this release excluding itself.

Watch the escape movie and download the original package. This repository contains the character-sheet dataset, not H3 video training clips.