scenario-reve

Scenario Reve Image

Overview

Reve, Reve AI's image family on Scenario, splits into two members that share a brand and little else: v2.1 is a reasoning-driven generator and instruction editor (a top-2 text-to-image arena entry at authoring time), Remix a fast, cheap compositor that blends several images into one. Discover them with search and treat model_schema_get as the contract: they disagree on field names, prompt caps, and ratio lists.

Connection and the core loop: see the scenario skill in this repo; model-agnostic image work: the scenario-image skill. If a sibling skill named here is missing from your available skills, ask the user to install it (npx skills add scenario-labs/skills --skill <name>); unattended, proceed from tool schemas and flag the gap.

Quick reference

At authoring time (the live schema wins):

Member	Capabilities	References	Prompt cap	Aspect ratios
Reve v2.1	txt2img, img2img	references, optional	4000	18 values, 4:1 through 1:4, default auto
Reve Remix	img2img	referenceImages, required, 1 to 6	2560	7 values, 16:9 through 9:16, default 3:2

Three parameters each: the prompt, the reference array, aspectRatio. Neither member has a mask, seed, width and height, negative prompt, or batch-count field, so an edit is scoped by wording and size comes only as a ratio. auto (v2.1 only) picks a fitting shape; pass an explicit ratio when the deliverable demands one.

Pick by job. v2.1 covers generation from a prompt alone and precise edits: swap the text on a poster, recolor a product, change a material, relight or restyle, merge references while every subject stays consistent. Remix blends photos, illustrations, or design assets into one cohesive composite. Where either would do, dry_run both: at authoring time a v2.1 run cost several times a Remix run and took over a minute at p50 against Remix's twenty-odd seconds.

Frame tags wire v2.1 references

In a v2.1 prompt, a reference is addressed with a <frame>N</frame> tag, numbered from 0 in the order the assets appear in references: the first is <frame>0</frame>, the second <frame>1</frame>. Use the tags whenever more than one reference is in play; without them the model guesses which image you mean. Remix documents no tag syntax: name subjects in plain words ("the bottle from the first image, the lighting from the second").

Edits are instructions, not masks

Neither member takes a mask, so the prompt carries the whole edit: state what changes and what must stay untouched. v2.1's reasoning rewards specific plain-language instructions over keyword lists, and its 4000-character budget leaves room for constraints. For masked inpainting on an exact region, use a model that exposes a mask field (see scenario-image).

Worked example: swapping the text on a poster

1. search with target="models", query="reve", public=true. At authoring time the hits were model_reve-v2-1 (generation and instruction edits) and model_reve-remix (multi-image blends); re-discover each session.
2. model_schema_get on the pick: field names, caps, and the ratio list before anything else.
3. upload_asset the poster (see the scenario skill) to get an asset id.
4. model_run with that model_id, dry_run=true, and parameters={"prompt": "Replace the headline on the poster in <frame>0</frame> with 'SUMMER SALE', matching the original font, perspective, and lighting. Change nothing else.", "references": ["asset_x"], "aspectRatio": "auto"} for the cost estimate.
5. Repeat model_run with wait=false, then jobs_wait with the returned job id, re-called with pending_job_ids on timeout, never a second model_run.
6. asset_display to check the swap, then asset_download to save.

Common mistakes

• Counting frames from 1: <frame>1</frame> is the second reference; the first is <frame>0</frame>.
• Carrying a field name across members: v2.1 takes references, Remix takes referenceImages; each schema lists only its own.
• Running Remix without references: referenceImages is required (1 to 6 at authoring time); prompt-only generation belongs to v2.1.
• Reusing a v2.1 prompt on Remix without checking length: 4000 characters overflow its 2560 cap.
• Sending width, height, a mask, or a seed: none exists on either member; aspectRatio is the only size control.
• Treating a v2.1 jobs_wait timeout as an error: p50 latency topped a minute at authoring time; re-call with pending_job_ids.