dreamina-canvas-discover-models

Dreamina Canvas Discovery

This Skill teaches every other dreamina-canvas-* Skill how to ask the CLI what is currently supported. It exists so that no caller hard-codes model names, voice names, ratios, resolutions, durations, or batch limits.

Reference detail is in references/capability-discovery.md.

When to use

• Before the first generation in any workflow that takes a --model, --voice-name, --ratio, --resolution, --duration, or --count.
• Before any audio music generation (must confirm the active environment's audio model and that the caller has not cached a name from a different environment).
• When a generation fails with "model not found" or a --voice-name rejection: re-discover rather than retrying with a remembered value.

When not to use

• As a substitute for dreamina-canvas-cli argv construction rules.
• To bypass a --credit-ceiling decision; discovery tells you the legal parameter values, it never approves spend.

Select commands from the live schema

Run version and schema before building discovery argv. If schema declares model search, use model search --detail full. If it declares model list instead, use model list; that compatibility shape already returns per-mode flags, enums, bounds, and reference requirements.

Never send an optional flag merely because a guide mentions it. Pass voice list --language <code> only when the live schema declares it; public 1.0.0 accepts only --offset and --count for voice list.

Discovery commands

# Summary catalog (one row per model)
dreamina-canvas --format json model search --type image

# Compatibility shape (only when the live schema declares model list)
dreamina-canvas --format json model list --type image

# Full spec per model (every legal flag, enum, bound, batch limit)
dreamina-canvas --format json model search --type image --detail full

# Single model full spec when you already have a name
dreamina-canvas --format json model <model-id> --type image

# TTS voice catalog (add --language only when the live schema declares it)
dreamina-canvas --format json voice list --offset 0 --count 50

# Argument spec for any subcommand (for the exact flag names)
dreamina-canvas --format json schema "node create image"

Use only the model command shape returned by schema. With model search, run summary discovery and then --detail full; with model list, read its full per-mode specifications directly. Never go from a remembered model name to generation: the catalog moves.

What to read from the discovery payload

The live full payload (model search --detail full or model list, as declared by schema) is the only authoritative source for:

• --model identifiers and aliases
• --ratio legal values per model
• --resolution legal values per model (some require it explicitly)
• --duration (video, music)
• --count upper bound, taken from generation.maxBatchGenCount
• Required references, VIP / entitlement flags (generation.isVip, vipConfigs[]), per-model constraints

For audio:

• TTS --voice-name must come from voice list or from model search --type audio --detail full. The CLI resolves the public voice name into the server-side authoritative identifier.
• Music --model must come from model search --type audio --detail full filtered to MODE=music. Music does not use a server-implicit default; the CLI refuses --model missing with cli.audio_music_model_required and exit code 2.

Hard rule: never cache across environments

The discovery payload is per environment (distribution × edition × profile). A model name valid in one environment is not portable to another.

• Re-run model search --type image --detail full whenever the active profile or environment changes (dreamina-canvas version reports distribution and edition; trust them).
• Re-run discovery after any CLI upgrade that bumps commit/buildTime.

What this Skill will not do

• Recommend a specific model, ratio, resolution, or voice by name.
• Treat a remembered name as equivalent to a re-discovery.
• Save the discovery payload to a public log, fixture, or example. Use placeholders like <model> / <voice-name> / <ratio> instead.

Policy

Invocation requires:

• Confirmed installation of dreamina-canvas
• Confirmed active profile and environment via dreamina-canvas version

Forbids:

• Hard-coding or recommending a specific model name, voice name, ratio, resolution, duration, or batch count
• Treating a discovery payload from one environment as valid in another
• Saving raw discovery payloads into public logs, fixtures, or examples

Default prompt:

Before generation, inspect schema, then use its declared full model discovery shape (model search --detail full or model list) and voice list for TTS. Add optional flags only when the live schema declares them.