Pangram's production AI image and video detector (released Nov 4, 2026): upload a picture or clip and it returns an AI-likelihood score plus a heatmap that pinpoints which regions look synthetic, with a faster video model, Pangram Image Flash 1.0, that analyzes clips frame by frame.
Last verified · 2026-11-06 · by Moe Ameen
Pangram Image 1.0 is an AI image detector released on November 4, 2026 by Pangram, the New York detection company founded by Stanford AI graduates Max Spero and Bradley Emi (the same firm whose text detector powers Substack's reader-facing AI scan). It is the production version of the AI image detector Pangram first shipped as a research preview earlier in the year. You upload a picture and it returns a probability that the image is AI-generated, together with a heatmap that localizes where the synthetic content sits — so it can flag an AI object or face composited into an otherwise real photo rather than scoring the whole file as one number.
Alongside it, Pangram released Pangram Image Flash 1.0, a smaller, faster model built for video. It analyzes an uploaded clip frame by frame, so you can see which parts of a video are flagged rather than getting a single verdict for the whole thing. Pangram reports 99.96% accuracy for Image 1.0 on its own evaluation set — 1,200 synthetic images built by mirroring 100 real photos with 12 frontier generators — and 98.7% for Flash 1.0 on the DF26 video benchmark, which it frames as a 29-point lead over the next-best detector. Those are company-reported figures; Pangram also cites academic benchmarks such as Synthbuster, Mirage, and Community Forensics.
Two things changed with this release that matter in practice. The minimum input size dropped from 512×512 to 256×256 pixels (results under 512 carry a reduced-confidence note), and the model is trained for robustness to compression, edits, blur, and noise — which makes it usable on the recompressed images that actually circulate on social feeds, not just clean exports. You reach it through Pangram's web app (image detection and video upload), a browser extension that flags AI content as you browse, and an API that covers both image and video.
The boundary to design around: Pangram Image 1.0 is a measurement instrument. It tells you how an image or clip reads; it creates nothing, brands nothing, and publishes nothing. Everything after the result — producing on-brand visuals, captioning, reframing, scheduling, publishing — is a separate job.
The most practical place to put Pangram Image 1.0 in a creator's stack is at the *inbound* edge — vetting the visuals that feed your content before you build anything on them. Kompozy is a content generation and publishing engine that ingests sources (uploads, RSS, webhooks) and turns them into finished video, images, carousels, blogs, and posts. If a stock image, a piece of UGC, a reaction clip, or a "found" photo you were about to repurpose is quietly AI-generated and undisclosed, you don't want to discover that after it's been reframed into a branded carousel and fanned across your feeds. Running incoming visuals through Pangram's heatmap first — and clips through Flash 1.0's frame-by-frame view — tells you what you're actually building on, so your Kompozy output rests on vetted source material.
From there the division of labor is clean: Pangram verifies, Kompozy produces. Instead of repurposing a questionable found image, generate the net-new asset in Kompozy — a Gemini face-locked Persona Photo, a HeyGen avatar Persona Short, a brand-exact HyperFrames carousel, or a Quote Graphic — held to a Persona Brief so it's unmistakably yours, then reframe it to 9:16, 1:1, and 16:9 and publish across eight social platforms plus blog and email. Pangram is the gate that keeps unknown synthetic media out of your pipeline; Kompozy is the engine that fills the pipeline with visuals you made on purpose.
It is Pangram's production AI image detector, released November 4, 2026. You upload an image and it returns a likelihood it is AI-generated plus a heatmap showing where the synthetic content sits. It ships with Pangram Image Flash 1.0, a faster model that analyzes video frame by frame, and is reachable via web app, a browser extension, and an API.
Pangram reports 99.96% accuracy on its own evaluation set of 1,200 synthetic images and 98.7% for Flash 1.0 on the DF26 video benchmark, plus strong results on academic benchmarks like Synthbuster and Mirage. These are company-reported figures on a fast-moving problem, so treat any single result as an estimate rather than proof.
Yes, through Pangram Image Flash 1.0, a smaller model released alongside it that analyzes an uploaded clip frame by frame so you can see which parts are flagged. Pangram reports it at 98.7% on the DF26 benchmark. Image 1.0 itself handles still images, with a minimum input size lowered to 256×256 pixels.
Use it as an inbound filter. Before repurposing a found image or clip, scan it with Pangram and read the heatmap; if it's undisclosed AI or ambiguous, generate the asset net-new in Kompozy instead — a face-locked Persona Photo, avatar video, or a brand-exact carousel — held to your Persona Brief, then publish across platforms. Pangram vets the inputs; Kompozy makes the output.