What are the key takeaways from this a16z Podcast episode?

Key insights include: **Model efficiency via domain focus:** Ideogram's open-weights model achieves competitive image quality at 9.3 billion parameters versus the previous state-of-the-art at roughly 80 billion — a 9x reduction — by concentrating on graphic design, typography accuracy, and layout control rather than competing on raw scale against compute-rich labs like Google.; **JSON as intermediate representation:** Ideogram routes prompts through a structured JSON format containing thousands of words per image, specifying every element, bounding box, and position before passing to the diffusion model. This gives professional users direct visibility into model inputs, enabling precise element-level edits and consistent output across iterations — unlike closed systems from OpenAI or Google.; **Enterprise customization workflow:** Brands consistently reject generic models because they fail to match style guidelines and brand DNA. Ideogram offers three customization tiers: open-source fine-tuning on the quantized model, a self-serve upload tool at $60/month requiring a minimum of 15 images, and a full enterprise engagement where Ideogram's annotation team curates training data with the client's design team.

What did Mohammed Nourouzi discuss on a16z Podcast?

Ideogram CEO Mohammed Nourouzi joins a16z's Yoko Lee and Justine Moore to discuss the company's first open-weights image generation model at 9.3 billion parameters, its JSON-based prompting architecture, design-focused differentiation strategy, and customization pathways for artists and enterprise customers. Key topics include: **Model efficiency via domain focus:** Ideogram's open-weights model achieves competitive image quality at 9.3 billion parameters versus the previous state-of-the-art at roughly 80 billion — a 9x reduction — by concentrating on graphic design, typography accuracy, and layout control rather than competing on raw scale against compute-rich labs like Google.; **JSON as intermediate representation:** Ideogram routes prompts through a structured JSON format containing thousands of words per image, specifying every element, bounding box, and position before passing to the diffusion model. This gives professional users direct visibility into model inputs, enabling precise element-level edits and consistent output across iterations — unlike closed systems from OpenAI or Google..

How long is this episode of a16z Podcast?

This episode is 42 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

a16z Podcast

AI, Design, and the Power of Open Models

June 15, 2026

42 min episode · 2 min read

Mohammed Nourouzi

Episode

42 min

Read time

2 min

Topics

Productivity, Fundraising & VC, Leadership

AI-Generated Summary

Published Jun 16, 2026

Key Takeaways

✓Model efficiency via domain focus: Ideogram's open-weights model achieves competitive image quality at 9.3 billion parameters versus the previous state-of-the-art at roughly 80 billion — a 9x reduction — by concentrating on graphic design, typography accuracy, and layout control rather than competing on raw scale against compute-rich labs like Google.
✓JSON as intermediate representation: Ideogram routes prompts through a structured JSON format containing thousands of words per image, specifying every element, bounding box, and position before passing to the diffusion model. This gives professional users direct visibility into model inputs, enabling precise element-level edits and consistent output across iterations — unlike closed systems from OpenAI or Google.
✓Enterprise customization workflow: Brands consistently reject generic models because they fail to match style guidelines and brand DNA. Ideogram offers three customization tiers: open-source fine-tuning on the quantized model, a self-serve upload tool at $60/month requiring a minimum of 15 images, and a full enterprise engagement where Ideogram's annotation team curates training data with the client's design team.
✓Editable design over flat image output: The next unreleased capability Nourouzi prioritizes is editable text and layout control — generating layered, modifiable design assets rather than single flat images. This directly addresses marketing and design workflows where teams need to adjust typography, reposition elements, or swap copy without regenerating entire compositions from scratch.
✓Small model enables on-device and artist workflows: Running on a single consumer GPU, the 9.3B parameter model opens two underserved segments: privacy-sensitive enterprises wanting on-premise deployment, and individual artists who can fine-tune the model to their personal style, texture, and canvas characteristics. One artist in residence reported a 3x speed increase producing a comic book using a customized version.

What It Covers

Ideogram CEO Mohammed Nourouzi joins a16z's Yoko Lee and Justine Moore to discuss the company's first open-weights image generation model at 9.3 billion parameters, its JSON-based prompting architecture, design-focused differentiation strategy, and customization pathways for artists and enterprise customers.

Key Questions Answered

•Model efficiency via domain focus: Ideogram's open-weights model achieves competitive image quality at 9.3 billion parameters versus the previous state-of-the-art at roughly 80 billion — a 9x reduction — by concentrating on graphic design, typography accuracy, and layout control rather than competing on raw scale against compute-rich labs like Google.
•JSON as intermediate representation: Ideogram routes prompts through a structured JSON format containing thousands of words per image, specifying every element, bounding box, and position before passing to the diffusion model. This gives professional users direct visibility into model inputs, enabling precise element-level edits and consistent output across iterations — unlike closed systems from OpenAI or Google.
•Enterprise customization workflow: Brands consistently reject generic models because they fail to match style guidelines and brand DNA. Ideogram offers three customization tiers: open-source fine-tuning on the quantized model, a self-serve upload tool at $60/month requiring a minimum of 15 images, and a full enterprise engagement where Ideogram's annotation team curates training data with the client's design team.
•Editable design over flat image output: The next unreleased capability Nourouzi prioritizes is editable text and layout control — generating layered, modifiable design assets rather than single flat images. This directly addresses marketing and design workflows where teams need to adjust typography, reposition elements, or swap copy without regenerating entire compositions from scratch.
•Small model enables on-device and artist workflows: Running on a single consumer GPU, the 9.3B parameter model opens two underserved segments: privacy-sensitive enterprises wanting on-premise deployment, and individual artists who can fine-tune the model to their personal style, texture, and canvas characteristics. One artist in residence reported a 3x speed increase producing a comic book using a customized version.

Notable Moment

Nourouzi reveals that Ideogram deliberately avoided reinforcement learning training on this model, which is the opposite approach of most frontier competitors. This keeps the model stylistically raw and diverse, allowing many distinct visual styles rather than converging on a single polished aesthetic that dominates current high-leaderboard models.

Know someone who'd find this useful?

Episode Transcript

It's not about how good a model is in the general sense. It's about how good is this model for my use case. For a lot of design and marketing use cases, we need editable design, not a single flat image. It's super impressive, honestly, reaching the level of things like nano banana or GPT image with an open source model. Why did you think that was important? We really want our models to have taste. Every artist, they can really customize this model to the nuances of their style, the texture of their canvas, and really get two k output, and hopefully make that part of their workflow. One thing we were always wondering is that this release open source model is so small. It's 9,300,000,000 parameters. Like, previously, the SOTA is probably, like, 80,000,000,000 parameters. It's, like, nine x of a difference. How did you do it? We focused That includes That includes everything from typography and layouts to editing, customization, and workflows that fit into professional creative processes. Yoca Lee and Justine Moore speak with Ideogram founder and CEO, Mohammed Nourouzi, about image generation, open weight models, design tools, and the future of creative AI. So today, we're excited to have Mohammed, CEO and founder of Ideogram, a Toronto based generative AI company that just released their first OpenWeights image model. Congrats on the huge release. Congratulations. Thanks for having me. We're really excited to talk through something that everyone has been buzzing about, which is the fact that the model is OpenWeights. The previous ITOGRAM models have been closed source, so we'd love to hear how you made the call to make it open this time. What has happened is there has been a lot of progress in industry, and we used to do everything. Basically, we had our own first party app as well as our own first party API, and model development itself is a lot of work. And we decided to focus a little more on the model side. We think that's where a lot of potential exists. We still want to continue to own the interaction with the users. We think there's a lot of important feedback we can get from the users directly, but then we wanna focus more on building the model. And by releasing the weights, we're actually extending ourselves and working with inference providers, working more directly with large enterprise. They have every ability to customize the models or host it on prem or optimize it for device. And we would love to work with the best chipmakers to really optimize the model, the best inference providers. So this is basically us saying, hey. We are very serious about building the foundation model, and we would like to work with you whoever you are, whether you're app developer or a chip maker or an inference provider. I think you already kind of touched on this. The new open source model is very exciting in that it unlocked a …

Get the full transcript (7,140 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all a16z Podcast transcripts →