upscale-image-workflow

Upscale Image Workflow

Ideogram exposes one tool for raising an image's resolution: upscale_image. This skill orchestrates a single call to that tool — resolve which image, confirm the settings, run it, save the output's identifiers. It takes exactly one call per invocation: no multi-step pipeline, no multi-tool surface. It's the thinnest skill in the repo.

This is an orchestration workflow around one MCP call, not a prompt-composition skill — there's no composition-spec-format.md/panel-anatomy.md-style reference here.

Before you start

Read references/upscale-settings.md before running step 1 or step 2 of the workflow below. It covers the identifier-resolution rule (upscale_image requires exactly one of image_response_id or image_upload_id — never both, never neither) and the settings guide for every optional parameter.

Workflow

1. Resolve the target image

Determine exactly one identifier to pass. Either:

• image_response_id — from this session's generate_image/remix_image/edit_image/ reframe_image/remove_background calls' structured_content.response_ids, from get_recent_generations's per-item response_id, or from get_images_by_collection_id's per-asset response_id/image_id.
• image_upload_id — from upload_image, for a local file the user points to that hasn't been uploaded yet. Call upload_image first rather than inventing an ID: per upload_image's own description, run its returned instructions curl command in the sandbox and read the real id back from the curl response.

If more than one candidate image is plausibly "that image" (e.g. several renders made earlier this session with nothing distinguishing which one the user means), ask which one — don't guess. Never pass both identifiers or neither. See references/upscale-settings.md for the full identifier-resolution rule.

2. Confirm settings

State the resolved settings before calling:

• upscale_factor — the user's stated value if given (that counts as confirmed, no need to re-ask), or the tool's own X2 default, stated plainly rather than blocked on a confirmation question.
• upscale_details_weight — only set if the user asked for more/less added detail; otherwise omit for the backend default.
• prompt — only set if the user wants to steer what detail gets added; otherwise omit — the backend auto-describes the image when none is supplied, per the tool's own description.
• collection_id — only if the user asked to save directly into an existing collection. If so, follow collections-management's find-or-create pattern from collection-patterns.md to resolve which collection_id to pass, rather than inventing one.
• private — omit unless the user explicitly asks to publish to the public Ideogram feed. Per the tool's own description, paid accounts default to private and free/Basic accounts default to public, and enterprise generations are always private regardless — state that as the only asserted fact about the default, since it's the tool's own documented behavior.