Je teste la nouvelle IA RunwayML Gen-3 — plus fort que Sora ?

Je teste la nouvelle IA RunwayML Gen-3 — plus fort que Sora ?

by Ludo Salenne

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQFGenerative artificial intelligence
🎙 Ludo Salenne 👥 267K 📅 July 7, 2024 ⏱ 28 min 👁 23K 📄 review 🧭 2026-08-21
Available in: English (current) Français

Keywords

RunwayMLGen-3AI videotestSora

Summary

In this video, Ludo Salenne tests RunwayML’s new Gen-3 Alpha model, which generates videos from text prompts. He subscribes to the service live, paying $144 for a yearly plan, and explores the interface. He attempts to recreate promotional videos using the same prompts, but the results are often disappointing, with issues like anatomical errors and lack of realism. He discovers that credits are consumed at 10 credits per second of video, and that the quality varies significantly. He then creates a custom GPT using ChatGPT to generate optimized prompts based on Runway’s official guide, and tests it with mixed results. The video also covers features like lip sync and the ability to reuse settings. Overall, the creator expresses skepticism about the hype, noting a gap between promotional demos and actual user results.

132 words

Critical Evaluation

Value of the Information & Strength of the Argument

The video provides valuable practical insights into using RunwayML Gen-3, including real-time subscription, credit costs, and prompt engineering. The creator’s argumentation is based on direct testing and comparison with official examples, which adds credibility. However, the evaluation is subjective and limited to a few examples, and the creator’s frustration may influence the narrative. The video also demonstrates a workflow for creating a custom GPT to improve prompt quality, which is a useful technique for viewers.

Scientific Rigor, Source Quality, Title Accuracy

The video does not cite scientific sources but references RunwayML’s official guide and examples. The creator is transparent about the limitations of the tool and the gap between marketing and reality. The title is accurate, as the video directly tests Gen-3 and compares it to Sora, though Sora is not available for direct comparison. The video includes a sponsored segment for the creator’s own training course, which is disclosed but not relevant to the content.

165 words

Title / Content Match

The title accurately reflects the content: the creator tests RunwayML Gen-3 and compares it to Sora, though the comparison is based on expectations rather than direct access to Sora.

Quality & Reliability

6/10

The video is a hands-on test of RunwayML Gen-3, providing practical observations and comparisons with promotional material. The creator is transparent about limitations and costs, but the analysis is subjective and based on a single user experience. No external scientific sources are cited.

Chapters

Cited Sources

  • RunwayML Gen-3 Alpha — Official product page for Gen-3 Alpha, used for prompts and examples.
  • Formation ChatGPT — Creator's own training course, mentioned as a resource.
  • Ressources IA — Free resources including prompts and GPTs, mentioned for viewers.

Concurring Sources

  • RunwayML Gen-3 Alpha — Official page showing examples and capabilities, which the video compares against.

Dissenting Sources

  • Sora by OpenAI — Sora is not yet publicly available, so the comparison is based on expectations rather than direct testing.

External References

Contribution & Novelties

The video offers a practical, hands-on evaluation of RunwayML Gen-3, highlighting the discrepancy between promotional demos and real-world results. It also introduces a workflow for creating a custom GPT to generate optimized prompts, which is a novel approach for viewers.

Pour aller plus loin :

  • RunwayML Gen-3 Alpha — Official product page for further details.
  • Sora by OpenAI — OpenAI’s text-to-video model, mentioned as a comparison.
  • Custom GPTs — OpenAI’s official blog on GPTs, relevant to the custom GPT creation.

80 words

Radar Profile

The radar profile shows moderate scores across all dimensions, with a slight emphasis on quantity of information and technical level. This reflects a video that provides practical details but lacks deep scientific rigor and authoritative sources.

Reliability 5/10

💬 No comments were provided for analysis.