---
title: 'Hierarchical Text-Conditional Image Generation with CLIP Latents'
license: CC-BY-4.0
license_url: https://creativecommons.org/licenses/by/4.0/
license_note: 'paper is CC BY 4.0 (verified on arXiv abs page 2026-11-12; icon by-4.0.png, http://creativecommons.org/licenses/by/4.0/); note is our own summary'
source: https://iinventedlight.com/library/notes/2204-06125-dalle-2.md
source_license: 'CC BY 4.0'
attribution: 'Summary by iinventedlight (AI-assisted). Source: Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, Mark Chen, Hierarchical Text-Conditional Image Generation with CLIP Latents (CC BY 4.0).'
author: 'iinventedlight (AI-assisted)'
date: 2026-11-12
status: published
approved_by: "Cole Withington, 2026-11-12 14:44 PT, Librarian chat widget t51s316: \"Yes, publish all 100 second-list arXiv notes\""
checked_on: 2026-11-12
as_of: 2026-11-12
confidence: medium
untested: 'Not reproduced. Abs page and HTML key sections read 2026-11-12. Named code URLs checked for HTTP 200; code not run.'
category: arxiv
topic_area: 'ai-research: multimodal'
tags: arxiv, multimodal, text-to-image, diffusion
arxiv_id: 2204.06125
alias: 2204.06125-dalle-2
sources:
  - id: S1
    title: 'Hierarchical Text-Conditional Image Generation with CLIP Latents'
    publisher: arXiv
    authors: 'Aditya Ramesh, Prafulla Dhariwal, Alex Nichol, Casey Chu, Mark Chen'
    url: https://arxiv.org/abs/2204.06125
    version_read: 'v1 (13 Apr 2022)'
    accessed: 2026-11-12
    license: 'CC BY 4.0'
---
# Hierarchical Text-Conditional Image Generation with CLIP Latents

## Summary
This paper proposes a two-stage text-conditional image generator that uses CLIP latents: a prior that generates a CLIP image embedding from a text caption, and a decoder that generates an image conditioned on that embedding. The authors show that explicitly generating image representations improves image diversity with minimal loss in photorealism and caption similarity. They use diffusion models for the decoder and report that diffusion priors are computationally more efficient and produce higher-quality samples than autoregressive priors. [S1]

## Claims
- id: C1
  claim: 'The method has a prior that generates a CLIP image embedding from a text caption, then a decoder that generates an image.'
  source: S1
  evidence: "a prior that generates a CLIP image embedding given a text caption"
- id: C2
  claim: 'The authors say explicitly generating image representations improves diversity with minimal loss in photorealism and caption similarity.'
  source: S1
  evidence: "improves image diversity with minimal loss in photorealism and caption similarity"
- id: C3
  claim: 'They find diffusion priors more efficient and higher-quality than autoregressive priors.'
  source: S1
  evidence: "the latter are computationally more efficient and produce higher-quality samples"

## Verification
- Code available: No official training-code URL was named on the abs page or HTML. A preview system card at https://github.com/openai/dalle-2-preview/blob/main/system-card.md 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](https://iinventedlight.com/library/notes/2103-00020-clip/). This batch also has the earlier zero-shot text-to-image (DALL-E) note.
- Limits: This v1 paper (13 Apr 2022) presents the CLIP-latent hierarchical method often discussed as DALL-E 2 / unCLIP. It is not a production-system card.
- 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 13 Apr 2022 (v1). Sources: [S1] https://arxiv.org/abs/2204.06125, v1 (13 Apr 2022). checked_on: 2026-11-12.

## When to apply
Use when asking how CLIP latents were used for hierarchical text-conditional image generation.

## Facts
- id: F1
  claim: 'The two-stage model generates a CLIP image embedding from text, then decodes an image from that embedding.'
  confidence: medium
  date: 2026-11-12
  sources: [S1]
  license: CC BY 4.0
- id: F2
  claim: 'The authors report improved image diversity with minimal loss in photorealism and caption similarity.'
  confidence: medium
  date: 2026-11-12
  sources: [S1]
  license: CC BY 4.0
- id: F3
  claim: 'They use a diffusion decoder and report that a diffusion prior is more efficient and higher quality than an autoregressive prior.'
  confidence: medium
  date: 2026-11-12
  sources: [S1]
  license: CC BY 4.0

## Sources
- [S1] arXiv 2204.06125, https://arxiv.org/abs/2204.06125. License: CC BY 4.0. v1 (13 Apr 2022). 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.

---

Agents: [/agents.md](/agents.md) | [/.well-known/agent-library.json](/.well-known/agent-library.json).
