Scenario Image Editing
Overview
Editing an existing image is a model_run on a tool model: one file in, a few numeric knobs, one or
more assets out, nothing to prompt on most of them. Generating a new image, prompt-driven edits and
masked inpainting are scenario-image; the same effects on footage are scenario-video-editing;
stacking layers is scenario-video-assembly (Image Studio). Connection and the core loop:
scenario. If a sibling skill named here is missing from your available skills, load it with
skill_get (Scenario MCP) or ask the user to install it
(https://github.com/scenario-labs/skills#install); unattended, proceed from tool schemas and flag
the gap.
Quick reference
Two discovery lanes. A family is browsed with search (target="models", public=true) by tag. A
single utility is a capability, so it goes to recommend with capability="img2img" and the need
in the user's own words, which names the purpose-built tool and prices it against the general
editors; when it answers a utility job with general editors only, find the tool by name with
search and filters={"tags": ["tool"]}, as the watermark row does:
| Need |
Route |
| The effects family |
search, filters={"tags": ["Post Processing"]} (40 hits, image and video) |
| Upscale or enhance |
search, filters={"tags": ["image-upscale"]} |
| Cutouts, relight, layers, vector |
recommend: at authoring time each lane returned its purpose-built tool ranked by measured cost and latency against the general editors, where search returned the same tools with nothing to rank them by |
| Watermarks, burned-in text |
search, query="text remover", a name lookup: at authoring time recommend sent watermark removal to general instruction editors and missed the purpose-built Photoroom Text Remover |
| Tiles, sheets |
none: model_scenario-image-slicer, model_scenario-grid-maker (fixed first-party ids, each Scenario's single deterministic tool for its job) |
Reframe ids below are authoring-time hits: re-discover them.
The effects family
Eighteen effects share one shape, a required image file plus one to three knobs: blur, chromatic
aberration, color correction, crystallize, cubism, desaturate, dissolve, dodge and burn, glow and
bloom, grain, 3D color LUT, oilify, parabolize, posterize, sharpen, solarize, tint, vignette.
Ranges are per tool and unguessable. posterizeThreshold runs 0 to 1 (default 0.5) while Color
Correction's temperature, contrast and saturation run -100 to 100 with gamma on 0.2 to 2.2,
all from model_schema_get. Not every knob is a number: Grain takes a 22-value profile enum, a
color temperature and a boolean, no strength control at all. Those profiles are looks, not
intensities, and some soften instead of texturing: compare against the input rather than assuming
grain landed. Grain's grainColorTemp (2000 to 10000) hides a sharper trap: the 6500 default is not
neutral, and it warms the frame and lifts blacks harder than a restrained LUT pass does, so the
texture step quietly re-grades what the grade step just set. Set it deliberately and judge the
result against the graded input, not the original; which direction neutralizes it is unverified, so
compare one frame each side of 6500 before a batch. Defaults disagree too: Color Correction's are
no-ops (nothing set returns the input unchanged and still charges) where LUT and Posterize ship a
visible default.
lutStyle holds 140+ exact strings, one of which contains a space (cgc_look_teal and orange), so
copy them from the schema rather than retyping. Prefixes group them: cgc_film_emulation_* and
rec709_* emulate film stocks, cgc_log_to_rec709_* expects log footage and will over-contrast an
ordinary render, and the bulk of the list (cgc_look_*, pond5_*, distant_land_*,
shutterstock_*) are look packs. Five bare presets sit outside every prefix, and one of them,
teal_orange, is the model's default: leave lutStyle unset and the grade that lands is teal and
orange, not neutral.
These finish inside model_run, returning status: "success" with the assets attached, so no
jobs_wait. They are flat-priced: at authoring time every effect dry-ran at 1 CU whatever the input
size (Resize Image at 2), so one dry_run stands for the family, where an upscale's price moves
with output pixels. Chain them by passing one run's asset_id to the next. The pipeline order
across this skill: reshape and upscale first (see the next section), then grade, then texture. Grain
and sharpening are high-frequency effects that any later resize interpolates away, so they go last,
at delivery resolution; a LUT is resolution-tolerant and sits on either side.
Upscaling is a model family, not a knob
Upscalers are img2img models, many taking no prompt at all: discover with
filters={"tags": ["image-upscale"]} (13 hits at authoring time). Fidelity upscalers (Topaz,
Recraft Crisp) sharpen and enlarge what exists, the pick when output must stay on-model; creative
ones (Magnific Creative, the Clarity pair) carry a creativity dial that invents detail and can
redraw fine features, so compare against the source. Sizing comes only from model_schema_get and
varies per model: a factor, a target resolution or megapixels, or just image with no dial (2x to
16x and 4K to 8K ceilings were typical, not bounds). Cost follows output pixels: dry_run=true,
top-level on model_run and never inside parameters, prices the exact size before a batch. An
upscale can also outrun model_run's wait budget where the effects above never do: a modest one
still returns status: "success" inline, a larger one returns status: "in_progress" and a job id
for jobs_wait, re-called with any returned pending_job_ids as job_ids. Purpose-built variants
protect seams on tileable textures and continuity on 360 panoramas: see scenario-textures and
scenario-skyboxes.
Expanding a canvas is four different tools
- Reframe (
model_scenario-gemini-reframe): an aspectRatio enum plus a resolution tier (1K,
2K, 4K), so exact pixels are out of reach; optional prompt. The enum is approximate: 4:5 came
back 1856x2304 (29:36), so a true ratio needs a Resize Image pass after it (fit: "cover", so it
crops the sliver instead of squashing).
- Smart Reframe (
model_scenario-smart-reframe): width and height are required and exact,
and it protects on-image text, brand marks and palette. textDensity: "DENSE" costs substantially
more.
- Photoroom Expand: exact
outputWidth and outputHeight up to 4096, plus a seed.
- Photoroom Uncrop: rebuilds a subject the frame edge cut off.
Resize Image (model_scenario-resize-image, a fixed first-party id: Scenario's single
exact-dimension resize tool, so discovery would only re-derive it) is none of these: it scales
pixels to width and height (one alone keeps the ratio) and never invents canvas; fit decides
what an off-ratio box does (contain, the default, may land smaller than the box, stretch
distorts, cover center-crops to exact dimensions).
Both reframes recompose generatively rather than filling canvas, and at tens of times an effect's
price they are the chain's expensive step: dry_run them. They re-render, so reframe first and
grade after: a grade or grain pass beforehand comes back partly reinterpreted.
Cardinality
Effects take a scalar image. Resize Image (images, max 10) and Grid Maker (images, max 100)
are array: true, where a bare id is silently dropped and the run succeeds having ignored it.
Worked example: grade a key art, then export at size
upload_asset the file, then upload_asset_complete unless it went inline under ~100KB.
- Reshape first: Resize Image to the delivery size, with
images as an array even for the one
file.
model_schema_get on model_scenario-postprocessing-lut (a fixed first-party id: Scenario's
single deterministic LUT tool, so discovery would only re-derive it), pick a lutStyle from its
enum, price it with model_run and dry_run=true, then run it with lutIntensity near 0.6 for
a restrained grade.
- Grain last, on that output, so its texture is sized for the shipping frame.
asset_display to review, asset_download to save.
Common mistakes
- Reaching for an
image_edit MCP tool, or for local ImageMagick or Pillow: the surface is
model_run on tool models.
- Prompting an effect ("more posterized"): they read numbers only. Reframe and the layer extractors
take text.
- Slicing or extracting layers before grading: both return one asset per piece, so every piece then
needs its own run.
- Saving a traced SVG with
asset_download: it comes back rasterized, so take the stored file from
asset_get's url with curl -L.
1---2name: scenario-image-editing3description: "Use when editing an existing image on Scenario through MCP with a tool model, not a new generation: upscale or enhance to 2x, 4K, 8K, super resolution, 3D LUT color grade, color correction, posterize, solarize, vignette, film grain, blur, sharpen, glow, chromatic aberration, oilify, cubism, crystallize, dodge and burn, tint, desaturate, expand or uncrop, reframe an aspect ratio, resize to exact pixels, slice tiles, contact sheet, split into layers, remove a background or watermark, vectorize."4license: MIT5metadata:6 source: https://skillmd.com/skills/scenario-labs/scenario-image-editing7---89# Scenario Image Editing1011## Overview1213Editing an existing image is a `model_run` on a tool model: one file in, a few numeric knobs, one or14more assets out, nothing to prompt on most of them. Generating a new image, prompt-driven edits and15masked inpainting are `scenario-image`; the same effects on footage are `scenario-video-editing`;16stacking layers is `scenario-video-assembly` (Image Studio). Connection and the core loop:17`scenario`. If a sibling skill named here is missing from your available skills, load it with18`skill_get` (Scenario MCP) or ask the user to install it19(https://github.com/scenario-labs/skills#install); unattended, proceed from tool schemas and flag20the gap.2122## Quick reference2324Two discovery lanes. A family is browsed with `search` (`target="models"`, `public=true`) by tag. A25single utility is a capability, so it goes to `recommend` with `capability="img2img"` and the need26in the user's own words, which names the purpose-built tool and prices it against the general27editors; when it answers a utility job with general editors only, find the tool by name with28`search` and `filters={"tags": ["tool"]}`, as the watermark row does:2930| Need | Route |31| -------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |32| The effects family | `search`, `filters={"tags": ["Post Processing"]}` (40 hits, image and video) |33| Upscale or enhance | `search`, `filters={"tags": ["image-upscale"]}` |34| Cutouts, relight, layers, vector | `recommend`: at authoring time each lane returned its purpose-built tool ranked by measured cost and latency against the general editors, where `search` returned the same tools with nothing to rank them by |35| Watermarks, burned-in text | `search`, `query="text remover"`, a name lookup: at authoring time `recommend` sent watermark removal to general instruction editors and missed the purpose-built Photoroom Text Remover |36| Tiles, sheets | none: `model_scenario-image-slicer`, `model_scenario-grid-maker` (fixed first-party ids, each Scenario's single deterministic tool for its job) |3738Reframe ids below are authoring-time hits: re-discover them.3940## The effects family4142Eighteen effects share one shape, a required `image` file plus one to three knobs: blur, chromatic43aberration, color correction, crystallize, cubism, desaturate, dissolve, dodge and burn, glow and44bloom, grain, 3D color LUT, oilify, parabolize, posterize, sharpen, solarize, tint, vignette.4546Ranges are per tool and unguessable. `posterizeThreshold` runs 0 to 1 (default 0.5) while Color47Correction's `temperature`, `contrast` and `saturation` run -100 to 100 with `gamma` on 0.2 to 2.2,48all from `model_schema_get`. Not every knob is a number: Grain takes a 22-value profile enum, a49color temperature and a boolean, no strength control at all. Those profiles are looks, not50intensities, and some soften instead of texturing: compare against the input rather than assuming51grain landed. Grain's `grainColorTemp` (2000 to 10000) hides a sharper trap: the 6500 default is not52neutral, and it warms the frame and lifts blacks harder than a restrained LUT pass does, so the53texture step quietly re-grades what the grade step just set. Set it deliberately and judge the54result against the graded input, not the original; which direction neutralizes it is unverified, so55compare one frame each side of 6500 before a batch. Defaults disagree too: Color Correction's are56no-ops (nothing set returns the input unchanged and still charges) where LUT and Posterize ship a57visible default.5859`lutStyle` holds 140+ exact strings, one of which contains a space (`cgc_look_teal and orange`), so60copy them from the schema rather than retyping. Prefixes group them: `cgc_film_emulation_*` and61`rec709_*` emulate film stocks, `cgc_log_to_rec709_*` expects log footage and will over-contrast an62ordinary render, and the bulk of the list (`cgc_look_*`, `pond5_*`, `distant_land_*`,63`shutterstock_*`) are look packs. Five bare presets sit outside every prefix, and one of them,64`teal_orange`, is the model's default: leave `lutStyle` unset and the grade that lands is teal and65orange, not neutral.6667These finish inside `model_run`, returning `status: "success"` with the assets attached, so no68`jobs_wait`. They are flat-priced: at authoring time every effect dry-ran at 1 CU whatever the input69size (Resize Image at 2), so one `dry_run` stands for the family, where an upscale's price moves70with output pixels. Chain them by passing one run's `asset_id` to the next. The pipeline order71across this skill: reshape and upscale first (see the next section), then grade, then texture. Grain72and sharpening are high-frequency effects that any later resize interpolates away, so they go last,73at delivery resolution; a LUT is resolution-tolerant and sits on either side.7475## Upscaling is a model family, not a knob7677Upscalers are `img2img` models, many taking no prompt at all: discover with78`filters={"tags": ["image-upscale"]}` (13 hits at authoring time). Fidelity upscalers (Topaz,79Recraft Crisp) sharpen and enlarge what exists, the pick when output must stay on-model; creative80ones (Magnific Creative, the Clarity pair) carry a creativity dial that invents detail and can81redraw fine features, so compare against the source. Sizing comes only from `model_schema_get` and82varies per model: a factor, a target resolution or megapixels, or just `image` with no dial (2x to8316x and 4K to 8K ceilings were typical, not bounds). Cost follows output pixels: `dry_run=true`,84top-level on `model_run` and never inside `parameters`, prices the exact size before a batch. An85upscale can also outrun `model_run`'s wait budget where the effects above never do: a modest one86still returns `status: "success"` inline, a larger one returns `status: "in_progress"` and a job id87for `jobs_wait`, re-called with any returned `pending_job_ids` as `job_ids`. Purpose-built variants88protect seams on tileable textures and continuity on 360 panoramas: see `scenario-textures` and89`scenario-skyboxes`.9091## Expanding a canvas is four different tools9293- **Reframe** (`model_scenario-gemini-reframe`): an `aspectRatio` enum plus a `resolution` tier (1K,94 2K, 4K), so exact pixels are out of reach; optional `prompt`. The enum is approximate: `4:5` came95 back 1856x2304 (29:36), so a true ratio needs a Resize Image pass after it (`fit: "cover"`, so it96 crops the sliver instead of squashing).97- **Smart Reframe** (`model_scenario-smart-reframe`): `width` and `height` are required and exact,98 and it protects on-image text, brand marks and palette. `textDensity: "DENSE"` costs substantially99 more.100- **Photoroom Expand**: exact `outputWidth` and `outputHeight` up to 4096, plus a `seed`.101- **Photoroom Uncrop**: rebuilds a subject the frame edge cut off.102103Resize Image (`model_scenario-resize-image`, a fixed first-party id: Scenario's single104exact-dimension resize tool, so discovery would only re-derive it) is none of these: it scales105pixels to `width` and `height` (one alone keeps the ratio) and never invents canvas; `fit` decides106what an off-ratio box does (`contain`, the default, may land smaller than the box, `stretch`107distorts, `cover` center-crops to exact dimensions).108109Both reframes recompose generatively rather than filling canvas, and at tens of times an effect's110price they are the chain's expensive step: `dry_run` them. They re-render, so reframe first and111grade after: a grade or grain pass beforehand comes back partly reinterpreted.112113## Cardinality114115Effects take a scalar `image`. Resize Image (`images`, max 10) and Grid Maker (`images`, max 100)116are `array: true`, where a bare id is silently dropped and the run succeeds having ignored it.117118## Worked example: grade a key art, then export at size1191201. `upload_asset` the file, then `upload_asset_complete` unless it went inline under ~100KB.1212. Reshape first: Resize Image to the delivery size, with `images` as an array even for the one122 file.1233. `model_schema_get` on `model_scenario-postprocessing-lut` (a fixed first-party id: Scenario's124 single deterministic LUT tool, so discovery would only re-derive it), pick a `lutStyle` from its125 enum, price it with `model_run` and `dry_run=true`, then run it with `lutIntensity` near 0.6 for126 a restrained grade.1274. Grain last, on that output, so its texture is sized for the shipping frame.1285. `asset_display` to review, `asset_download` to save.129130## Common mistakes131132- Reaching for an `image_edit` MCP tool, or for local ImageMagick or Pillow: the surface is133 `model_run` on tool models.134- Prompting an effect ("more posterized"): they read numbers only. Reframe and the layer extractors135 take text.136- Slicing or extracting layers before grading: both return one asset per piece, so every piece then137 needs its own run.138- Saving a traced SVG with `asset_download`: it comes back rasterized, so take the stored file from139 `asset_get`'s `url` with `curl -L`.