Use image URLs for image understanding prompts
Provide public internet URLs for images instead of base64 data URLs when sending images to models.
The image_url value can also be a public URL on the Internet instead of a base64 data URL.
Explanation
This allows models to consider image context when generating responses, similar to text-only prompts, by referencing publicly accessible images.
Example
Instead of embedding a base64 string, use a URL like 'https://example.com/image.jpg'.
What it does and when to use it
When sending images to models for understanding, use public internet URLs for the image source. This method allows the model to incorporate the image's context into its generated response, functioning similarly to a text-only prompt.
Sources and credit
- Image Understanding ↗xAI · xAI Docs: Guides · May 29, 2026 · paraphrased
First seen on Subvolts May 29, 2026. Updated Sep 11, 2026. Spotted a problem? Tell us.
Where it shows up
- Image UnderstandingxAI · xAI Docs: Guides · May 29, 2026
Related in the PlaybookPlaybook →
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- Anti-patternStoring request/response history with image inputsAvoid storing request/response history on the server when sending images to prevent potential request failures. · 5 min
- ToolMMPostTrainBench Benchmark for Autonomous Multimodal AI ResearchBenchmark for evaluating AI agents in sustained multimodal AI research and model improvement. · 10 min
- WorkflowAutomated synthetic map and route generation pipelineGenerate diverse, high-quality maps with annotated paths using AI models. · 120 min
- Model pickMedSigLIP for medical image classification and retrievalUse MedSigLIP for medical imaging tasks requiring structured outputs like classification or retrieval. · 5 min
- HeuristicOrchestrate Multiple Tools with Image GenerationCombine image generation with other tools like web search in a single agentic loop. · 5 min
- Troubleshooting fixImage Edit Fails to Change Detail: Point to the PartWhen an image edit doesn't work, specify the exact part of the image to change. · 5 min
Questions people ask
- Which AI tools does this work with?
- Any current model. It was seen with Grok in the sources.
- Where did this come from?
- xAI (xAI Docs: Guides). Subvolts collected, normalized and credited it; the link above opens the original.