# Search images, video, audio or page images with text

> Search images, video, audio or page images with text: 0 companies fully do it, 11 with a limit, 35 carry the tag. Every listing in Embeddings & rerankers with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/embeddings.multimodal-search/
- Markdown: https://www.anchorterminal.com/niches/embeddings.multimodal-search/index.md (~1,800 tokens)
- Slim: https://www.anchorterminal.com/niches/embeddings.multimodal-search/index.min.md (~430 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/embeddings.multimodal-search/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 0 | 11 | 35 | 12 | 11 |

## What counts as done

Text and media embed into one vector space through a documented call, per-request caps (images, seconds of video or audio, PDF pages) are published, longer media is segmented by the API or by documented guidance, and the docs say whether a video's audio track is embedded.

Measured, for our own tests: text-to-image and text-to-video recall@10 on a labelled set, and cost per hour of video.

How agents ask for it: find the clip where a car runs the red light; search my photos for team dinner; find the slide with this chart; search the video archive by what is said or shown

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Amazon Nova Multimodal Embeddings (Amazon Web Services, Inc.)**: Text, images, document images, video and audio share one space; sync caps are 30 seconds, async 2 hours with API segmentation by durationSeconds, and AUDIO_VIDEO_COMBINED or SEPARATE says how a soundtrack is embedded. Served only in us-east-1, a named blocker.
- **Cohere Embed and Rerank (Cohere Inc.)**: Text, images and mixed text-and-image inputs such as PDF pages share one space, but there is no video or audio input, and image calls are limited to 5 a minute on trial keys.
- **EmbeddingGemma 2 (Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author))**: Text, images, video and audio share one 768-dimension space with per-modality caps (about 29 images, 58 frames, 327 seconds of audio), but no segmentation guidance for longer media, and the card doesn't say whether a video's soundtrack is embedded.
- **Gemini Embedding (Google LLC)**: One space for text, images, video, audio and PDF with caps (6 images, 120 s video, 180 s audio, one 6-page PDF), and video audio tracks are stated as not processed; no segmentation guidance for longer media, and the Vertex SDK dropped image parts with a 400.
- **Jina Embeddings and Reranker (Jina AI GmbH)**: v5-omni takes text, images, video, audio and PDFs in one call shape, but the reference gives no seconds or frame caps for video or audio, no segmentation, and doesn't say whether a video's soundtrack is embedded.
- **jina-embeddings-v5-omni-small (Jina AI GmbH, a subsidiary of Elastic N.V. (per the jina-embeddings listing's 2 October 2026 check))**: Text, images, video and audio in one space, and the card says a video's soundtrack is never read unless passed as its own part; no per-input caps in seconds or frames, no segmentation guidance, and 8-minute loads reported.
- **Nomic Embed (Nomic, Inc.)**: nomic-embed-vision-v1.5 embeds images for search against text embeddings, but there is no video or audio and no per-request image cap on the pages read.
- **NVIDIA NeMo Retriever Embedding and Reranking NIMs (NVIDIA Corporation)**: llama-nemotron-embed-vl-1b-v2 and Nemotron 3.5 embed text, images and text with images (page images) with a 5 MiB image cap, but there is no video or audio input.
- **Qwen3-VL-Embedding-2B (Alibaba Cloud (RDAP names Alibaba Cloud (Singapore) Private Limited as the registrant of qwen.ai))**: Text, images, screenshots and video share one space with MMEB-V2 video results, but no per-input frame or pixel caps, no segmentation guidance, no audio, and an account found vLLM vectors didn't match the Transformers path.
- **Qwen3-VL-Embedding-8B (Alibaba Group (Tongyi Lab, named in the copyright line of the GitHub repository's Apache 2.0 licence))**: Text, images, screenshots and video share one space with MMEB-V2 video results, and an account added offline dashcam video search with it, but no per-input caps, no segmentation guidance and no audio.
- **Voyage AI embeddings and rerankers (Voyage AI Innovations, Inc.)**: voyage-multimodal-3.5 embeds text, images and video in one space with caps (16 million pixels, 20 MB per image or video, 32,000 tokens per input), but no audio, no word on a video's soundtrack, no segmentation guidance, and inputs train models by default.

- not this job: EmbeddingGemma 300M (Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author)): Text input only.
- not this job: Granite Embedding Small English R2 (International Business Machines Corporation): Text only.
- not this job: Mistral Embed and Codestral Embed (Mistral AI (RCS Paris 952 418 325)): Text and code input only.
- not this job: OpenAI embeddings (OpenAI OpCo, LLC): Text input only.
- not this job: Qwen3-Embedding-0.6B (Alibaba Cloud (the registrant of qwen.ai per RDAP, carried from the Qwen3.8-27B check)): Text input only.
- not this job: Qwen3-Embedding-4B (Alibaba Cloud (named in the copyright line of the 8B sibling's Apache 2.0 licence; this repository has no LICENSE file)): Text input only.
- not this job: Qwen3-Embedding-8B (Alibaba Cloud (named in the licence's copyright line)): Text input only.
- not this job: ZeroEntropy zerank and zembed (ZeroEntropy, Inc.): Text models only.

## What stops agents finishing

- SDK filters out non-text parts and the API returns 400 (Gemini Embedding 2 on Vertex)
- runtime preprocessing differs from the reference, so vectors from two routes do not match (Qwen3-VL in vLLM)
- uploading private footage to a hosted API is not acceptable to the user
- model served in one cloud region only (Nova, us-east-1)
- load or embed time for video makes the job impractical on a laptop

## Around it

Before: file-storage.list-files, speech-to-text.long-recording-batch. After: vector-search.document-results, local-ai.screen-and-meeting-recall.

## Where this comes from

Weight 12 of 100 in Embeddings & rerankers: 11 of 62 accounts (91 account-use-case tags, accounts counted in each use case they fit, scaled to 100 and rounded by hand); dashcam, photo, screen-history and card-image search on Show HN, plus the EmbeddingGemma 2 launch thread and Gemini and Qwen3-VL issues

As data: https://www.anchorterminal.com/api/v1/use-cases/embeddings.multimodal-search.json, or the MCP tool get_use_case with id embeddings.multimodal-search. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
