Claude Code image generation · Updated November 16, 2026
Can Claude generate images? Not on its own. Here's how to add it to Claude Code.
Claude doesn't create photos or illustrations. It can draw diagrams and simple graphics as SVG or HTML, and it can read images you give it. If you need an actual image file, Claude Code has to call an image model. With UseTrawsn that takes one command, and the file lands in your project.
What Claude can and can't do with images
Anthropic's help article says Claude doesn't generate photos or illustrations, and its model docs list text as the only output. Here's how that plays out for common requests.
| What you ask for | Claude on its own | Claude Code + UseTrawsn |
|---|---|---|
| Photo, illustration or product shot | No. Claude has no image model. | An image model such as Seedream 5, GPT Image 2 or Nano Banana Pro makes it, and the file is saved where you point -o. |
| Icon, diagram or exact interface labels | Yes, as SVG or HTML code that Claude Code writes in the project. | Same route. No image model or credits needed. |
| Edit an existing image | Only scripted edits like crop, resize or filters, if Claude can run code. | Image-to-image edits, using the saved file as the reference. Unchanged areas aren't guaranteed to stay identical. |
| Understand an image or screenshot | Yes. Claude reads JPEG, PNG, GIF and WebP images. | Same as before. No UseTrawsn call needed. |
This workflow runs in the Claude Code terminal environment. Installing the skill here does not configure the Claude chat website or another host. The recorded lamp example below uses the CLI directly; it is not a test of automatic skill selection.
Your Claude subscription and UseTrawsn generation usage are separate. The example cost 2 credits for the original and 2 for the edit; check current pricing and your UseTrawsn balance before a new request. Installing instructions alone does not provide model access.
Checked November 2, 2026 against Anthropic's Can Claude produce images?, models overview and upload documentation.
A real generation and edit, with one small artifact
On September 22, 2026, we ran Seedream 5 through UseTrawsn CLI 0.6.2 against the production API. The first request generated the olive lamp; the second used that saved file as its image reference. Both returned 2848 × 1600 PNGs with the requested 2k setting. The CLI reported 2 credits per request and command wall times of 28.24 and 33.22 seconds, including transfer and download. These are single-run observations, not fixed prices or latency guarantees.


The yellow shade and top cap are visible in the revision, while the stem, base, and overall composition remain similar. Inspection also found a small purple mark where the stem meets the base. This matters: asking to change only the shade did not leave every other pixel untouched. Both outputs are retained here as resized JPEG previews. The original PNGs were inspected directly and compared with UseTrawsn image analysis; this is one CLI workflow trial, not a benchmark of Claude Code itself.
Install path
Step 1
npx -y skills add usetrawsn-ai/usetrawsn -a claude-code -yStep 2
curl -fsSL https://usetrawsn.ai/install.sh | shStep 3
usetrawsn loginStep 1 adds the skill to Claude Code, step 2 installs the usetrawsn CLI, and step 3 signs you in. Then list the models with usetrawsn image models, run the text-to-image command below, or ask Claude Code for an image in plain language. Full options are in the Agent Skill setup, CLI reference and image docs.
First commands to try
Text to image
usetrawsn image generate --model seedream-5 --mode text-to-image --prompt "A small olive-green desk lamp on an off-white background, soft studio light, no text" --param aspect_ratio=16:9 --param resolution=2k --param format=png -o claude-code-launch-visual.pngImage editing
usetrawsn image generate --model seedream-5 --mode image-to-image --prompt "Change only the lamp shade to warm yellow. Keep the composition and background." --param images=./claude-code-launch-visual.png --param aspect_ratio=16:9 --param resolution=2k --param format=png -o claude-code-launch-visual-edited.pngFrom a project brief to an asset you can use
Try this request in Claude Code after setup: “Use UseTrawsn to create a 16:9 image of an olive desk lamp for this project. Show me the available models before generating, save the selected result locally, then make a second version with a yellow shade. Keep both files so I can compare them.” We ran the two CLI commands above for the sample; the surrounding Claude Code conversation is an example, not a recorded host test.
Define the destination
Specify the asset filename, aspect ratio, subject, and where it will appear. For exact interface labels or a simple diagram, SVG or HTML may be the more controllable choice.
Review before editing
Open the first file and inspect the subject, composition, unwanted text, and dimensions. Reference that saved image for the edit; a second text-only prompt does not preserve the first result.
Check before integrating
Compare both files. A request to change only one detail can still alter other areas. Keep the original and add the chosen asset to the project only after review.
Recover a failed image step before generating again
Authentication or model access fails
Run usetrawsn status in the same terminal where Claude Code executes commands. Resolve login, balance or model access there before trying another generation; credentials in a different host session do not establish access here.
The reference image is rejected
Confirm the saved file exists and inspect the image-to-image schema. Pass it as --param images=./claude-code-launch-visual.png so the CLI can upload it. The Seedream 5 schema checked on October 12 accepts up to nine references; check the selected model instead of reusing another model's limits.
An output was returned but the local save failed
Keep the returned output URL and request or task ID. Retry that download to a writable path before creating another image. If the submission result is unknown, investigate the recorded request instead of automatically resubmitting a billed call.
usetrawsn image models seedream-5 schema --operation generate --mode image-to-imageusetrawsn download "<returned-output-url>" -o claude-code-launch-visual.pngAgent skill or direct CLI
Agent skill
Best when you want Claude Code to choose the appropriate UseTrawsn command from a natural-language request and keep the result inside the current task.
Direct CLI
Best when you want explicit control over the model, mode, input image, parameters, and output filename. Claude Code can run the same command for you.
Current image models Claude Code can call
- GPT Image 2
- OpenAI's image family for general generation and prompt-driven image edits.
- Seedream 5
- The model used in the examples below: text-to-image and image-to-image, with 2k or 4k output in the checked schema.
- Nano Banana Pro
- Supports generation and reference-image editing. Inspect its own schema before reusing parameters from another model.
- Nano Banana 2
- Another generation and editing option. Compare results and actual cost on your own brief before choosing for repeated runs.
Catalog and command check · Catalog and CLI rechecked October 21, 2026
Choose a model and inspect its input requirements before generating.
Claude Code supplies the planning and terminal orchestration; UseTrawsn supplies the image runtime. The production catalog check exposes thirteen active image models and both text-to-image and image-to-image modes. Generated files can be downloaded directly into the working project with -o. This check confirms model discovery and request fields; it is not an end-to-end generation benchmark.
- CLI checked
- UseTrawsn CLI 0.6.5
- Live catalog
- 13 active image models
- Catalog modes
- text-to-image, image-to-image
- Delivery
- Local output path with -o
usetrawsn image models seedream-5 schema --operation generate --mode image-to-imageCost and limits to check first
Generation uses your UseTrawsn account and is separate from a Claude subscription. In the September 22, 2026 run above, each Seedream 5 request was charged 2 credits. At the Starter pack rate ($10 for 500 credits, $0.02 per credit, checked on UseTrawsn pricing on November 16, 2026), that is about $0.04 per image, or $0.08 for the generate-plus-edit pair. That is a single observation, not a quoted price: the CLI reports the actual charge with each successful request, it depends on the model and settings, and revisions add requests. The checked Seedream 5 image-editing schema accepts up to nine references; other models can have different limits. Local references are uploaded for processing, and the command needs permission to write the output file. An authentication error must be resolved before generation can succeed.
FAQ
Does Claude generate images in the Claude app?
Not photos or illustrations. In the web and desktop apps, Claude can build diagrams, charts, and interactive visuals from HTML and SVG (Anthropic labels this beta). The UseTrawsn setup on this page runs in Claude Code and other terminal agents, not in the claude.ai chat.
Is this a Claude Code skill?
Yes. UseTrawsn ships an agent skill for Claude Code. The skill teaches Claude Code when and how to use the UseTrawsn CLI for image generation and editing.
How do I install the UseTrawsn image generation skill in Claude Code?
Three commands. Run npx -y skills add usetrawsn-ai/usetrawsn -a claude-code -y to add the skill, curl -fsSL https://usetrawsn.ai/install.sh | sh to install the UseTrawsn CLI binary, then usetrawsn login. After that the skill is ready inside any Claude Code session.
Can Claude Code make images once UseTrawsn is installed?
Yes. Claude Code still writes SVG or HTML on its own; for photos and illustrations it calls an image model through UseTrawsn. The skill only tells Claude Code what to do. You also need the CLI, an UseTrawsn login, and access to the model you pick.
Which image models does the skill expose?
The October 21, 2026 production catalog check lists thirteen active image models. Examples include GPT Image 2.5 Flare and Sunburst, Seedream 5, and the Nano Banana family. Run usetrawsn image models and inspect the selected model schema before production use; the catalog can change.
Do I need a different workflow for image editing?
No. The same usetrawsn image command in the skill covers both text-to-image and image-to-image workflows depending on the model and flags you use.
Generate one image in Claude Code
One saved raster image. Review the example and edit workflow below before expanding into a batch.
Install the CLI and sign in once. If you are already configured, continue with the access and cost checks.
curl -fsSL https://usetrawsn.ai/install.sh | sh
usetrawsn login
usetrawsn --version
usetrawsn statusAdd the skill, then reopen your agent session if needed.
npx -y skills add usetrawsn-ai/usetrawsn -a claude-code -yRun usetrawsn status, inspect the current model schema, and check pricing and your balance. Generation is billed; model access and rates can change.
Paste this prompt into your configured agent. Inspect the delivered file and actual charge.
Use UseTrawsn to generate a small olive-green desk lamp on an off-white background, soft studio light, no text. Check my login, image catalog and Seedream 5 text-to-image schema. Save the result as ./desk-lamp.png. Explain the applicable pricing and ask for approval before the billed task. After success, report the saved file and actual charge; do not count a submitted or failed request as a delivered image.Success means the output file exists and can be opened. Compare it with your brief or reference, including dimensions, subject details and unwanted text. If generation returned an output URL but saving failed, retry downloading that output before requesting another image.
Read the related workflow. For setup help, see installation and login.
CLI version and updating · 0.6.5
Current release checked October 21, 2026: 0.6.5. Use the update command for your installation method, then verify the version and login in the same environment that runs your agent.
Direct binary installation
usetrawsn updatenpm installation
npm install -g @usetrawsn/cli@latestusetrawsn --version
usetrawsn statusVersion 0.6.5 retries transient downloads, recovers an expired video link by querying the same task, and saves multiple image outputs to numbered files. After a video download failure, resume the existing task instead of generating again.
usetrawsn video tasks wait <task-id> -o result.mp4Dated examples retain the CLI version, output and charge from the original run. Updating the CLI does not revalidate an old result. Check the live model schema and current pricing before a new billed task.