Adobe Resize Photos and Videos

Resizes images or videos to user-specified dimensions, aspect ratio, or named size. For social media platform sets (Instagram, TikTok, etc.) → use the adobe-create-social-variations skill instead.

---

Tool Reference

Step	Tool	Notes
Open file picker	asset_add_file	Returns presignedAssetUrl, assetId, and mediaType
Find asset in CC	asset_search	Re-stage result via asset_add_file for image; use id directly for video
Render PSD/AI	document_render_vector	Converts to JPEG before resize
Resize image	image_crop_and_resize	One call per dimension; reframe by default
Preview outputs	asset_preview_file	All outputs in a single call
Upload to CC (start)	asset_initialize_file_upload	First call in block upload pipeline
Upload to CC (finish)	asset_finalize_file_upload	Second call; returns presigned CC URL for board
Create Firefly board	create_firefly_board	Pass import_adobe_storage with finalize URL
Resize video	video_resize	Returns statusId for polling
Poll video resize	resizeVideoPoll	Deferred tool — load before calling

---

Workflow

Step 0 — Initialize Adobe Tools

Call adobe_mandatory_init first. This returns file handling rules and tool routing guidance required for the rest of the workflow.

{ "skill_name": "adobe-resize-photos-and-videos", "skill_version": "1.0.1" }

---

Step 1 — Entitlement Check

Now that adobe_mandatory_init confirmed that the "Adobe for creativity" connector is live, check which tools are available through the "Adobe for creativity" connector by cross checking against the Tool Reference table above.

---

IMMEDIATE ACTION REQUIRED

Step 1: If parameters are missing or ambiguous → run the multi-step intake (Step 2 below). Step 2: If no file is uploaded, run Step 3 (asset_add_file({})). Step 3: Detect image vs. video and route to the correct workflow (Step 5 Image/Video).

---

Step 2 — Multi-Step Intake

Run this when the user hasn't provided complete parameters (type + dimensions + fit mode). Skip entirely if the user's message already contains all three.

Use AskUserQuestion to collect parameters step by step.

Step 2a — Type

AskUserQuestion([{
  question: "Photo or video?",
  type: "single_select",
  options: ["Photo", "Video"]
}])

Step 2b — Dimensions

If Photo (multi_select — user can pick one or more):

AskUserQuestion([{
  question: "Dimensions — select all you need",
  type: "multi_select",
  options: [
    "1080 × 1080 — Instagram square",
    "1080 × 1350 — Instagram portrait",
    "1080 × 1920 — Stories / Reels",
    "1920 × 1080 — YouTube / landscape",
    "1280 × 720 — YouTube thumbnail",
    "1200 × 675 — X / Twitter post",
    "1200 × 630 — Facebook / OG image",
    "1080 × 566 — LinkedIn post",
    "2560 × 1440 — YouTube banner",
    "3840 × 2160 — 4K / UHD",
    "400 × 400 — Profile / avatar",
    "2480 × 3508 — A4 print (300 dpi)",
    "Custom — I'll type it"
  ]
}])

If user includes "Custom" → follow up: "Enter your custom dimensions (e.g. 1500×800):"

If user selects multiple presets → run image_crop_and_resize once per dimension, using the same source URI and parameters each time. Preview all outputs together at the end.

If Video (multi_select — user can pick one or more):

AskUserQuestion([{
  question: "Dimensions — select all you need",
  type: "multi_select",
  options: [
    "1080 × 1920 — Reels / TikTok / Stories",
    "1920 × 1080 — YouTube / landscape",
    "1080 × 1080 — Square post",
    "1280 × 720 — YouTube 720p",
    "3840 × 2160 — 4K / UHD",
    "854 × 480 — SD / 480p",
    "Custom — I'll type it"
  ]
}])

If user selects multiple video presets → submit a separate video_resize job per dimension, poll each to completion, preview all outputs together.

Step 2c — Fit mode

If Photo (single_select):

AskUserQuestion([{
  question: "Fit mode",
  type: "single_select",
  options: [
    "Reframe — crop to fill",
    "Pad — letterbox, no crop",
    "Extract — isolate subject"
  ]
}])

If Video (single_select):

AskUserQuestion([{
  question: "Fit mode",
  type: "single_select",
  options: [
    "Letterbox — preserve ratio",
    "Crop — fill, no bars",
    "Stretch — force exact (distort)"
  ]
}])

Step 2d — Smart focus (Photo only)

AskUserQuestion([{
  question: "Smart focus — where should the crop center?",
  type: "single_select",
  options: [
    "Subject",
    "Face / portrait",
    "Upper body",
    "Center (geometric)",
    "Named object — I'll describe it"
  ]
}])

If user selects "Named object" → follow up: "Describe the object (e.g. 'the red car', 'the logo'):"

Step 2e — Options (Photo only)

AskUserQuestion([{
  question: "Any additional options?",
  type: "multi_select",
  options: [
    "Batch — multiple images",
    "Output: JPEG",
    "Output: PNG"
  ]
}])

Step 2f — Source (Video only)

AskUserQuestion([{
  question: "Where is the video?",
  type: "single_select",
  options: [
    "Upload from device",
    "From Creative Cloud"
  ]
}])

---

Step 3 — Get the Source File

Critical — /mnt/user-data/uploads/ paths are NOT usable URLs. Even when the user has attached a file and it appears in context at a /mnt/user-data/uploads/ path, you cannot pass this path to any Adobe tool. Call asset_add_file() to stage the file — /mnt/user-data/uploads/ paths are not fetchable by Adobe tools.

WRONG  ❌  imageURI: "/mnt/user-data/uploads/photo.jpg"
WRONG  ❌  imageURI: "file:///mnt/user-data/uploads/photo.jpg"
CORRECT ✅  call asset_add_file() → use presignedAssetUrl from picker context

No file yet → open the picker immediately:

asset_add_file()

Detect type from returned mediaType:

• image/* → IMAGE WORKFLOW
• video/* → VIDEO WORKFLOW

---

Step 4 — Resolve Target Dimensions

User says	Resolve to
1920×1080, 1920x1080	width: 1920, height: 1080
4K	3840 × 2160
HD, 1080p	1920 × 1080
720p	1280 × 720
480p	854 × 480
A4 print	2480 × 3508 (300 DPI portrait)
Letter print	2550 × 3300 (300 DPI portrait)
16:9, 1:1, 4:5, 9:16	ratio string → pass as output: "16:9"
square	output: "1:1"
Width only (800px wide)	width: 800, use ratio string to maintain aspect

If no target was specified in the form AND none in the user's message → ask:

"What dimensions do you need? (e.g. 1920×1080, A4, 16:9)"

---

IMAGE WORKFLOW

Step 5 (Image) — Valid URI Sources

Only these work as imageURI:

• presignedAssetUrl from asset_add_file() ✅
• outputUrl from a prior tool call ✅
• Public HTTPS URL provided by user ✅

Not usable as imageURI:

• /mnt/user-data/uploads/ paths ❌
• renditionURL from asset_search directly ❌
• file:// URIs ❌

Source	Resolution
Uploaded file	asset_add_file() → presignedAssetUrl
CC storage asset	asset_search → re-stage via asset_add_file() → presignedAssetUrl
Prior tool output	Use outputUrl directly
PSD / AI file	document_render_vector (JPEG, 300 DPI) → outputPresignedUrl

---

Step 6 (Image) — Fit Mode

Intent	Mode	Notes
"resize to WxH" / default	reframe	Largest crop at target ratio, centered — no distortion
"no cropping", "letterbox", "fit inside"	pad	Fills empty space with white
"isolate subject", "tight crop"	extract	Tight crop around detected region

Default: reframe. Change only when user is explicit.

⛔ fit: "extract" is incompatible with an explicit { width, height } — it stretches the image. Use a ratio string or omit output.

---

Step 7 (Image) — Call image_crop_and_resize

image_crop_and_resize({
  imageURI: presignedAssetUrl,
  options: {
    output: { width: 1920, height: 1080 },  // or "16:9"
    fit: "reframe",
    focus: "subject",  // or "face", "upper_body", { prompt: "the car" }, { x:0.5, y:0.5 }
    quality: 7
  },
  outputFileType: "jpeg"  // or "png" if transparency needed
})
// result.outputUrl → use for preview or chaining

Focus guide:

• "subject" — default; full-body detection, landscapes, objects
• "face" — portraits, headshots
• "upper_body" — chest-up crops
• { prompt: "the logo" } — named object from user's form input
• { x: 0.5, y: 0.5 } — geometric center (no clear subject)

---

Step 8 (Image) — Batch

If batch was selected in the form OR the user provided multiple files:

1. Call asset_add_file() for each file to get individual presigned URLs
2. Call image_crop_and_resize for each with the same parameters
3. Collect all outputUrl values
4. Call asset_preview_file with all outputs in a single call

Calls run sequentially in this environment. For >10 images, confirm with the user before starting.

Pagination across turns (batches >20): For large batches, the conversation context may fill before all images are processed, causing Claude to pause mid-batch. This is expected — not an error. When it happens:

• State how many have been completed and how many remain (e.g. "20 / 24 done — say 'continue' for the rest")
• Wait for the user to confirm before resuming
• On resume, pick up from the next unprocessed image and continue to the end
• Preview all outputs together only after the final image completes

---

Step 9 (Image) — Preview + Deliver + Firefly Board

asset_preview_file({
  assets: [
    { name: "photo_1920x1080.jpg", presignedAssetUrl: result.outputUrl },
    // all successful outputs
  ]
})

Save to CC + Create Firefly Board

After preview, upload each output to CC storage using the block upload pipeline (asset_initialize_file_upload → PUT chunk → asset_finalize_file_upload).

Then call the firefly board tool with the final output urls:

create_firefly_board({
  import_adobe_storage: [
    final_output_url_1,
    final_output_url_2,
    // ...
  ]
})

✅ Confirmed working (Stage): import_adobe_storage with presigned URLs from asset_finalize_file_upload successfully populates boards.

❌ Does not work: import_generic_assets with CC URNs (urn:aaid:sc:US:...) — USS indexing lag causes blank boards.

If board creation fails or the URL is malformed, omit the board link (retrying does not help).

Post the completion summary after the board step:

✅ Done! [N] image(s) resized and ready.

📥 Download:
• resized_1920x1080.jpg → <cc_storage_url>
[list individually if N ≤ 3, or single folder link if N > 3]

🎨 View in Firefly Board → <board_url>   ← omit if board creation failed

---

VIDEO WORKFLOW

Step 5 (Video) — Get Asset ID

Video tools require assetId, not a URL.

Source	How
Upload via asset_add_file	Use returned assetId
CC storage	asset_search with filters.mediaType: "video/mp4" → use asset id

---

Step 6 (Video) — Common Targets

Name	Dimensions
4K / UHD	3840 × 2160
1080p / HD	1920 × 1080
720p	1280 × 720
480p	854 × 480
Vertical 9:16	1080 × 1920
Square 1:1	1080 × 1080

---

Step 7 (Video) — Call video_resize

⚠️ resizeVideoPoll is a deferred tool — load it first before attempting to poll. Direct calls without loading will fail with "not loaded" error.

video_resize({
  assetId: videoAssetId,
  width: targetW,
  height: targetH,
  mode: "letterbox"  // default; see mode table below
})
// Returns { jobStatus: "pending", statusId }

Mode — use exact API values (not user-facing labels):

User says	API value	Behavior
"fit", "preserve ratio", "letterbox"	"letterbox"	Preserves ratio, adds black bars (default)
"fill", "crop to fill", "no bars"	"crop"	Crops edges to fill target exactly
"stretch", "force exact"	"stretch"	Distorts to exact target

ℹ️ Auto-reframe for video is not available in Claude yet, but new tools are being added all the time. If you need more control, try using Adobe Premiere.

Polling: load the deferred tool resizeVideoPoll before calling it — direct calls fail with "not loaded".

// Step 1: load the deferred tool resizeVideoPoll
// Step 2: poll until complete
resizeVideoPoll({ statusId })
// Repeat until jobStatus === "completed"
// Show progress % to user while waiting

---

Step 8 (Video) — Preview + Deliver + Firefly Board

If polling succeeds and returns outputUrl, upload to CC storage via the block upload pipeline (asset_initialize_file_upload → PUT → asset_finalize_file_upload) to get a presigned URL, then:

asset_preview_file({
  assets: [{ name: "resized_video.mp4", presignedAssetUrl: result.outputUrl }]
})

create_firefly_board({
  import_adobe_storage: [ presignedAssetUrl_from_finalize ]
})

✅ Pass import_adobe_storage with the finalize presigned URL. If board creation fails, omit the link.

✅ Done! Video resized to [WxH].

📥 Download → <cc_storage_url>

🎨 View in Firefly Board → <board_url>   ← omit if board creation failed

If polling is unavailable → skip preview, deliver the CC storage fallback message from Step 7.

---

Error Handling

Error	Action
/mnt/user-data/uploads/ or file:// passed as URI	Stop. Call asset_add_file() first.
Image too large after resize	Re-run with quality: 5
Video assetId not found	Re-upload via asset_add_file
Source too small for target	Warn user upscaling won't add detail; proceed if confirmed
Invalid video_resize mode	API only accepts "letterbox", "crop", or "stretch" — "fit" and "fill" will be rejected.
extract + {W,H} → stretching	Switch to reframe
No dimensions given	Render form or ask explicitly before calling any tool
Unsupported format	Ask user to export as JPG/PNG (images) or MP4 (video)
resizeVideoPoll not loaded / unavailable	Known gap in current connector. Skip polling. Tell user to check CC Files for output.

---

Common Dimension Reference

Use Case	Dimensions
Full HD / web hero	1920 × 1080
4K	3840 × 2160
Web thumbnail	1280 × 720
Square	1080 × 1080
A4 print (300 DPI)	2480 × 3508
US Letter print (300 DPI)	2550 × 3300
Blog featured image	1200 × 630
Email header	600 × 200
Avatar / profile	400 × 400
Instagram portrait	1080 × 1350
Stories / Vertical	1080 × 1920

---

Assets

assets/intake-form.html — Interactive intake form for resize parameters. Render at the start of any resize workflow where parameters are missing or ambiguous. Covers photo (target size, fit mode, smart focus, batch, output format) and video (target size, fit mode, source).