✕ Clear all filters
19 articles
▶ Videos →

📰 Dev.to · Elise Moreau

19 articles · Updated every 3 hours · View all reads

All Articles 127,673Blog Posts 132,510Tech Tutorials 32,953Research Papers 24,707News 18,118 ⚡ AI Lessons
Classifier-free guidance above 7.5 oversaturated our product renders
Dev.to · Elise Moreau 💻 AI-Assisted Coding ⚡ AI Lesson 3w ago
Classifier-free guidance above 7.5 oversaturated our product renders
TL;DR: Classifier-free guidance above a scale of ~7.5 pushed our SDXL product renders into...
View lesson → + Playlist
Top 5 LLM Gateways in 2026
Dev.to · Elise Moreau 🧠 Large Language Models ⚡ AI Lesson 3w ago
Top 5 LLM Gateways in 2026
As enterprise adoption of generative AI accelerates, teams are moving away from direct, hard-coded...
View lesson → + Playlist
Channels-last memory format cut our conv backbone latency 22%
Dev.to · Elise Moreau 📐 ML Fundamentals ⚡ AI Lesson 3w ago
Channels-last memory format cut our conv backbone latency 22%
TL;DR: Switching our convolutional segmentation backbone to PyTorch's channels-last memory format cut...
View lesson → + Playlist
A VLM gate for generated images, with provider failover via Bifrost
Dev.to · Elise Moreau 👁️ Computer Vision ⚡ AI Lesson 3w ago
A VLM gate for generated images, with provider failover via Bifrost
TL;DR: At Photoroom we run a vision-language model as the last check before a generated product image...
View lesson → + Playlist
The SDXL VAE overflow that decoded black images in fp16
Dev.to · Elise Moreau 📐 ML Fundamentals ⚡ AI Lesson 3w ago
The SDXL VAE overflow that decoded black images in fp16
TL;DR: The SDXL VAE decoder pushes activations past 65504, the max value fp16 can hold, so the last...
View lesson → + Playlist
Unifying image inputs across three vision providers behind Bifrost
Dev.to · Elise Moreau 🎨 Image & Video AI ⚡ AI Lesson 3w ago
Unifying image inputs across three vision providers behind Bifrost
TL;DR: We run an automated visual QA step that scores generated product shots with vision LLMs from...
View lesson → + Playlist
You Can't Govern the AI You Can't See
Dev.to · Elise Moreau 4w ago
You Can't Govern the AI You Can't See
AI governance starts with visibility: a policy, a budget, or a guardrail can only act on the AI...
Read article → + Playlist
The seam our tiled upscaler left on every 4K product render
Dev.to · Elise Moreau 1mo ago
The seam our tiled upscaler left on every 4K product render
TL;DR: We tile high-res images through our upscaler because a full 4096×4096 pass blows past 24GB of...
Read article → + Playlist
Centralising tool access for our prompt-assembly agent with Bifrost MCP gateway
Dev.to · Elise Moreau 1mo ago
Centralising tool access for our prompt-assembly agent with Bifrost MCP gateway
TL;DR: Before our SDXL stack renders a single product photo, a small LLM agent assembles the request...
Read article → + Playlist
Tracing our 4-stage product photo pipeline through Bifrost
Dev.to · Elise Moreau ☁️ DevOps & Cloud ⚡ AI Lesson 1mo ago
Tracing our 4-stage product photo pipeline through Bifrost
TL;DR: We added OpenTelemetry tracing across the four LLM and VLM hops in our product-photo pipeline...
View lesson → + Playlist
Semantic caching the VLM step in our product-photo pipeline
Dev.to · Elise Moreau 🛠️ AI Tools & Apps ⚡ AI Lesson 1mo ago
Semantic caching the VLM step in our product-photo pipeline
TL;DR: We put Bifrost in front of the VLM step that captions and rewrites prompts for our...
View lesson → + Playlist
The bf16 grad accumulator that killed our SDXL LoRA training
Dev.to · Elise Moreau 🧠 Large Language Models ⚡ AI Lesson 1mo ago
The bf16 grad accumulator that killed our SDXL LoRA training
TL;DR: Our SDXL LoRA fine-tune for a Photoroom product photography model trained for six days while...
View lesson → + Playlist
Cost accounting for diffusion image generation at $0.0008 per render
Dev.to · Elise Moreau 1mo ago
Cost accounting for diffusion image generation at $0.0008 per render
TL;DR: Per-image cost on our SDXL-based product photography pipeline at Photoroom dropped from...
Read article → + Playlist
Why your diffusion model is slow at batch size 1 (and what actually helps)
Dev.to · Elise Moreau 1mo ago
Why your diffusion model is slow at batch size 1 (and what actually helps)
TL;DR: Single-image diffusion inference is bottlenecked by kernel launch overhead and attention...
Read article → + Playlist
Routing diffusion inference traffic across three providers
Dev.to · Elise Moreau 1mo ago
Routing diffusion inference traffic across three providers
TL;DR: We route a mix of diffusion and LLM traffic across three providers from a single Go-based...
Read article → + Playlist
Why your diffusion model is slow at batch size 1 (and what actually helps)
Dev.to · Elise Moreau 📐 ML Fundamentals ⚡ AI Lesson 2mo ago
Why your diffusion model is slow at batch size 1 (and what actually helps)
TL;DR: Single-image diffusion inference is bottlenecked by kernel launch overhead and attention...
View lesson → + Playlist
Why Your Diffusion Model Is Slow at Inference (And It's Not the UNet)
Dev.to · Elise Moreau 📐 ML Fundamentals ⚡ AI Lesson 2mo ago
Why Your Diffusion Model Is Slow at Inference (And It's Not the UNet)
TL;DR: Most inference bottlenecks in diffusion pipelines are not in the UNet denoising loop. They are...
View lesson → + Playlist
Why Your Diffusion Model Is Slow at Inference (And It's Not the UNet)
Dev.to · Elise Moreau 📐 ML Fundamentals ⚡ AI Lesson 2mo ago
Why Your Diffusion Model Is Slow at Inference (And It's Not the UNet)
TL;DR: Most inference bottlenecks in diffusion pipelines are not in the UNet denoising loop. They are...
View lesson → + Playlist
Kimi K2.6 Is a Legit Opus 4.7 Replacement
Dev.to · Elise Moreau 🧠 Large Language Models ⚡ AI Lesson 2mo ago
Kimi K2.6 Is a Legit Opus 4.7 Replacement
For a long time, Opus 4.7 has been the default recommendation when someone asks for a top tier model....
View lesson → + Playlist