jimeng-image

# Jimeng Image Generation Use this skill when the task is specifically about **Jimeng image generation**. This skill assumes website interaction happens through the **OpenClaw-managed browser**. Prefer the built-in `browser` tool when available, otherwise use: ```bash openclaw browser --browser-profile <profile> <subcommand> ... ``` For the browser workflow itself, follow the shared `openclaw-browser` skill rules: - use an explicit profile whenever possible - prefer a dedicated profile for the current agent - re-snapshot whenever refs go stale - do not fall back to a generic unmanaged browser flow ## What this skill covers - open Jimeng image creation pages - enter prompts - set aspect ratio and generation options - submit image jobs - wait for outputs to finish - open the output detail page when needed - download local output files ## Preconditions Before doing anything: 1. Use a dedicated Jimeng-capable browser profile when possible. 2. Confirm the page is already logged in if login is required. 3. If Jimeng shows a login gate, approval prompt, payment gate, or exhausted credits, stop and tell the user clearly. Do **not** silently switch to another website or another image service. ## Primary entry URL ```text https://jimeng.jianying.com/ai-tool/home/?workspace=0&type=image ``` ## Profile policy Prefer this order: 1. current agent's dedicated browser profile 2. user-requested profile 3. configured OpenClaw default profile 4. `openclaw` fallback profile If a profile named after the current agent exists and is configured, prefer it. ## Standard operating pattern ### 1) Open the correct Jimeng page Open the image generation URL. ### 2) Take a fresh interactive snapshot Do this before each important interaction. ### 3) Fill the main prompt textbox Use the main creation textbox, not search boxes from the inspiration/discovery area. Observed Jimeng UI often contains: - a large prompt textbox for creation - other secondary textboxes such as discovery/search inputs Always identify the correct input from a fresh snapshot before typing. ### 4) Adjust generation settings if requested Possible settings include: - aspect ratio - model / generation mode - quality Use the actual live UI labels from the current snapshot. Do not assume fixed refs. ## Aspect ratio handling Jimeng usually exposes ratio through a visible button such as: - `16:9` - `1:1` - `9:16` - or an image-side ratio control such as `智能比例` ### Video ratio behavior observed in live testing Clicking the visible ratio button can expose a radio group with options like: - `21:9` - `16:9` - `4:3` - `1:1` - `3:4` - `9:16` When the user asks for a specific ratio: 1. click the ratio button 2. re-snapshot if needed 3. choose the requested radio option 4. re-snapshot again before generating If no ratio is specified, keep the default shown by Jimeng. ## Image generation workflow Typical flow: 1. open the image URL 2. snapshot 3. type the user's prompt into the main image prompt box 4. optionally set ratio / quality / model 5. click the generate button 6. wait for result cards or a result detail page 7. if needed, click a chosen result to open details 8. use `下载` when available to save the file locally ### Image result heuristics observed in live testing On the result/detail page, Jimeng may show buttons such as: - `下载` - `生成视频` - `智能超清` - `超清` - `细节修复` - `再次生成` If the user's request is simply "generate one image and give it to me", it is acceptable to: - choose the first completed result - open it - download it locally - report the saved file path If the user explicitly wants selection among multiple outputs, present the options instead of picking silently. ## Download rules Always download the final selected file to a local path. Prefer stable, descriptive names such as: - `jimeng-image-<timestamp>.png` After download: 1. confirm the file exists 2. report the absolute local path 3. if possible, attach or preview the media for the user ## Jimeng-specific UI hints Live testing on this environment showed common labels such as: - `图片生成` - `去查看` - `下载` - `再次生成` - `回到底部` These are useful anchors, but the exact layout and refs can change. Always trust the latest snapshot over memory. ## Failure handling ### If refs fail - take a new interactive snapshot - use the new refs - do not keep retrying stale refs ### If login is required - stop and tell the user Jimeng needs login - do not fake completion ### If credits or membership are insufficient - report exactly what the page shows - do not keep retrying blindly ### If generation is rejected or moderated - surface the reason if visible - ask whether the user wants a rewritten prompt ### If multiple outputs appear - if the user asked for just one result, choose the first clearly completed result - if quality comparison matters, ask before downloading all of them ## Anti-patterns Avoid these mistakes: - using a generic browser instead of OpenClaw browser - typing into the wrong textbox from the discovery/search area - assuming the generate button label is always visible - assuming the page auto-jumps to the finished result page - using stale refs after a generation refresh - forgetting to download the final file locally ## Minimal image checklist - open image URL - snapshot - type prompt - set requested ratio/options - generate - wait for result - open chosen result if needed - click `下载` - confirm local file path ## Success criteria A successful Jimeng image task should end with all of the following: - the requested image was actually generated on Jimeng - the requested ratio/options were applied when specified - the finished output was downloaded locally - the user receives the resulting media or the exact saved file path