Eliot Andres

Building something new. Prev: co-founder & CTO at @photoroom_app (YC S20)

[Eliot Andres @EliotAndres]

My favorite Claude prompt at the moment:

We added a lot of features to the app, custom logic for staging etc... List them all in an artifact with keep/cut and "Copy result", so I can decide what to keep what to remove

[Eliot Andres @EliotAndres]

@albn @cdricpinson @threejsconf It looks very useful and simple to use, congrats!

[Matt Rouif @matthieurouif]

Hot take: the AI image model you picked doesn't matter.

We benchmarked the best ones on 850 real products. Top performer kept every product detail intact 29% of the time. Logos and text warped in 1 in 5 outputs.

That's the ceiling. All of them. Right now.

[Eliot Andres @EliotAndres]

@AniC_dev I'm working at Wework Marais for the day and I can find a desk for you, in case you're around DM me

[Photoroom @photoroom_ML]

Reasoning made LLMs better. We think image editing is next. That's what we're building at Photoroom: the fidelity layer. Before generating, the model figures out exactly what has to be preserved. After generating, it checks its own output against that, and corrects itself, instead of trusting a single forward pass. Same bet as every reasoning model: don't trust one pass. Give the system room to think, verify, and fix its own work.

[apolinario (poli) @multimodalart]

Just dropped the first SenseNova U1.5 demo here ⬇️!
Still optimizing it, but wanted to release as fast as possible for folks to try

this is really nano banana 2 level, fully open source (Apache 2), try it and let me know what you think

[Banodoco @banodoco]

76 five-second clips exploring different animation styles with MiniMax H3.

All generated locally on a six-year-old GPU by the_shadow_nyc:

[Ahmed Dahbi @dahbiahmed]

unnecessary part 2 - the suit means strategy 🤷

[Photoroom @photoroom_ML]

Cold starts used to be why you’d avoid scale-to-zero for GPU inference. Not anymore, we got them down to seconds.

Built with the @nvidia Dynamo team using GPU memory checkpointing. Our engineer @david_lc98 wrote up what it really takes to run in production, gotchas included: https://t.co/Cac6JLbU9A

[Y Combinator @ycombinator]

AI is moving into the physical world.

We're excited about a new wave of startups rebuilding the systems that power the real world, from education and healthcare to defense, finance, infrastructure, and work itself.

https://t.co/QCIz6DnQnN

[Matt Rouif @matthieurouif]

Shadows are a new designer skill between artistic direction, brand and design.

At Photoroom, we built an AI Shadows model specifically for product. You can control intensity, softness, and direction, and apply it across 1, 20 or 1 million products in seconds.

This is what we mean when we say AI gives creative directors more control, not less. It also shows the future of product photography: high product fidelity, with most of the artistic and brand direction happening post-capture, on a computer, not in a studio.

[Matt Rouif @matthieurouif]

Today we are on the main stage of Startup School at Station F to talk about ambition for your startup with my co founder @EliotAndres , interviewed by @collinmathilde .

The reason we applied to @ycombinator is that we were both convinced it would increase our ambition and the impact of Photoroom on the world. We didn't know exactly how, but I had heard @dessaigne talk about the impact of YC: first aiming at a $100M valuation, then $100M in revenue.

The reality is you get to work with ambitious and smart people, and we all hold ourselves accountable. This ambition, with the right guidelines, actually creates unicorns. When you're in a group of 200 and 5% of them create unicorns, dozens of people have grown, and they're not necessarily smarter or more ambitious than you. It becomes obvious you have a shot. There's a part of luck, but a lot of it is drive. And the most important thing is being surrounded by ambitious people.

We were lucky to be surrounded by the YC community and other ambitious startups like Francescu, Jean from Mojo, and @julien_c from Hugging Face. I clearly remember Jean saying: "Matt, if you're the best in the world at removing backgrounds, this is already a $1B idea.". This turned out to be quite true and gave use some ambition shots.

And one thing that @gustaf 🇺🇦, our YC partner, always pushed us toward was extracting ambitious visions from all our conversations with users and our passion for helping them, while giving us shortcuts that YC alumni before us had proven to work.

You can build an ambitious startup with an HQ in Paris, but you need to surround yourself with ambitious, crazy people and always reconnect to the valley to set the bar higher.

[Corentin Kérisit @corentinanjuna]

I finally was able to compile and build a real iOS app from Linux, using @bazelbuild and distributed Linux workers.

I used @Dimillian's IceCube app, got Codex to modularize it to allow parallel Swift compile actions, and let it loose on 100s of cheap remote Linux workers!

[Eliot Andres @EliotAndres]

@NickFromIT @pkuhar Amazing, I didn't know! Thank you

[Peter Kuhar @pkuhar]

I didn't expect this to work out of the box.
Pretty awesome.

[Matt Rouif @matthieurouif]

We are so confident that @photoroom_ML has the highest product fidelity in the world for images that we guarantee it. Fidelity or token refunded.

Just walked off the Black Stage at VivaTech with an announcement I’ve wanted to make for a long time.

Photoroom now comes with a product fidelity guarantee: if our AI ever changes your actual product , your logo, your colors, the details that matter, you get your token back.

In ecommerce, fidelity isn’t optional. The customer has to receive exactly what they saw. What you see is what you get.

It only works because of two models behind the scenes, our foundation model research PRX Pixel , and a visual quality model that catches any drift before an image ships.

Also great to catch up on research with our long time advisors thanks to @ylecun and @julien_c . 🙏

[Max Brodeur-Urbas @MaxBrodeurUrbas]

"what if we spent our marketing budget on musicians instead"

[Linoy Tsaban @linoy_tsaban]

ICYM: a new 7B pixel space model landed 🔥

> created a prompt reactive demo to visualize the output of each step aligned with the changed prompt
I was curious to see what prompt changes would propagate well and how early on (my fox went from pan to panda, but the duck stayed pink)

Try it out👇

[Ostris @ostrisai]

AI Toolkit now supports training LoRAs on PRX Pixel.