Zero-Shot Text-to-Image Generation
Summary
This paper describes a simple approach to text-to-image generation based on a transformer that autoregressively models text and image tokens as a single stream of data. The authors contrast this with prior work that focused on more complex architectures, auxiliary losses, or side information such as object-part labels or segmentation masks. They report that, with sufficient data and scale, the approach is competitive with previous domain-specific models when evaluated in a zero-shot fashion. [S1]
Claims
- id: C1 claim: 'The approach is a transformer that autoregressively models text and image tokens as one stream.' source: S1 evidence: "autoregressively models the text and image tokens as a single stream of data"
- id: C2 claim: 'The authors evaluate the approach in a zero-shot fashion.' source: S1 evidence: "when evaluated in a zero-shot fashion"
- id: C3 claim: 'They say that with sufficient data and scale it is competitive with previous domain-specific models.' source: S1 evidence: "our approach is competitive with previous domain-specific models"
Verification
- Code available: Paper: https://github.com/openai/DALL-E, which returned HTTP 200 on 2026-11-12. Not run.
- Reproduced: not reproduced (2026-11-12). Code not run.
- Related work: Live note 2103-00020-clip. This batch also has the later CLIP-latent hierarchical generator note.
- Limits: This is the original zero-shot text-to-image transformer paper (v2, 26 Feb 2021). It does not describe the later CLIP-latent DALL-E 2 system. Competitiveness claims are author-reported on the paper's comparisons.
- License: CC BY 4.0 on the abs page (2026-11-12; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 24 Feb 2021 (v1); last revised 26 Feb 2021 (v2). Sources: [S1] https://arxiv.org/abs/2102.12092, v2 (26 Feb 2021). checked_on: 2026-11-12.
When to apply
Use when tracing the original DALL-E text-and-image token transformer, before later diffusion and CLIP-latent systems.
Facts
- id: F1 claim: 'The paper models text and image tokens autoregressively as a single stream in a transformer.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F2 claim: 'The authors evaluate the system in a zero-shot setting.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
- id: F3 claim: 'They report that with sufficient data and scale, it is competitive with previous domain-specific models.' confidence: medium date: 2026-11-12 sources: [S1] license: CC BY 4.0
Sources
- [S1] arXiv 2102.12092, https://arxiv.org/abs/2102.12092. License: CC BY 4.0. v2 (26 Feb 2021). Accessed 2026-11-12.
License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.
License: CC-BY-4.0 · Summary by iinventedlight (AI-assisted). Source: Aditya Ramesh, Mikhail Pavlov, Gabriel Goh, Scott Gray, Chelsea Voss, Alec Radford, Mark Chen, Ilya Sutskever, Zero-Shot Text-to-Image Generation (CC BY 4.0).
Cite this
'iinventedlight (AI-assisted)'. 2026. "Zero-Shot Text-to-Image Generation." iinventedlight Library, iil:note:jdqv0mtg, version of 2026-11-12. https://iinventedlight.com/library/notes/2102-12092-dalle/. License: CC-BY-4.0.
@misc{iil_note_jdqv0mtg,
author = {{'iinventedlight (AI-assisted)'}},
title = {{Zero-Shot Text-to-Image Generation}},
howpublished = {iinventedlight Library},
year = {2026},
date = {2026-11-12},
url = {https://iinventedlight.com/library/notes/2102-12092-dalle/},
note = {Permanent ID iil:note:jdqv0mtg. License: CC-BY-4.0.},
}
Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.
For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2102-12092-dalle.md.