#Built-in Image/OCR analyzer is failing

1 messages · Page 1 of 1 (latest)

indigo atlas
#

clawdbot tresponded with this today:

About your annotated screenshot: my image analyzer flaked mid-read, so I can’t reliably extract the redlines from the image alone. Best options:
•⁠  ⁠Paste the pseudo-code / annotations as text here, or
•⁠  ⁠Tell me the exact bullet list of changes you want for the profile page sections (even rough), and I’ll implement and iterate using live localhost screenshots.

clawdbot opened my browser, but unable to screenshot apparently? Anything i can do to ensure this works? MCP level or somethign?

tame geyserBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

tacit dewBOT
indigo atlas
#

furhter context!

[Richbot 🦁] Got the screenshots (and I can open them locally in the browser like I just did). One issue: my built-in image/OCR analyzer is currently failing (and the backup model is blocked by low Anthropic credits), so I can’t reliably “read” the redlined annotations from the images automatically. I can still use the images visually + compare against the live page on ⁠ localhost ⁠ and implement changes, but I’ll need you to tell me the intended changes in text (even short bullets), otherwise I might miss specifics.
#

Built-in Image/OCR analyzer is failing

tacit dewBOT
# indigo atlas furhter context! ``` [Richbot 🦁] Got the screenshots (and I can open them local...

Based on the error messages, the browser screenshotting is actually working — the issue is with the image analysis model that reads the screenshots.

The error says:

  • "image analyzer flaked mid-read"
  • "backup model is blocked by low Anthropic credits"

This means the image tool (which uses agents.defaults.imageModel) failed, likely because:

  1. Claude image analysis ran out of Anthropic credits
  2. The fallback model isn't configured or also failed

Solutions

Option 1: Add Anthropic credits (if you want to keep using Claude for image analysis)

Option 2: Configure a cheaper fallback image model in ~/.clawdbot/clawdbot.json:

# 
{
  agents: {
    defaults: {
      imageModel: {
        primary: "anthropic/claude-sonnet-4-5",  // or whatever you use
        fallbacks: [
          "openai/gpt-4o"  // cheaper for image analysis
        ]
      }
    }
  }
}

Option 3: Use the workaround — paste your annotated changes as text bullets, or describe them verbally. The agent can still implement changes by comparing the live localhost page.