Video Generation

Today's Stories

1. 1

   AMD Data Center jumps 107%, but Nvidia's $75B quarter widens the lead

   AMD posted record revenue of $11.54 billion with Data Center revenue of $6.72 billion, up 107% year over year. NVIDIA delivered $75 billion in total quarterly revenue with Data Center revenue up 92%. AMD's Helios rack—combining EPYC server chips, MI450 GPUs, and Pensando networking—drew strong customer interest, with Anthropic committing to up to two gigawatts of MI450 series GPUs, with the first gigawatt starting in 2027. NVIDIA's Blackwell is shipping into every major hyperscaler, and Vera Rubin production begins in Q3. AMD is mounting a genuine challenge to NVIDIA's dominance by offering hyperscalers a credible second source of gigawatt-scale AI compute—an attractive alternative to sole-vendor risk. However, NVIDIA maintains a structural advantage: its Data Center gross margin of 75% far exceeds AMD's 56%, and NVIDIA's vertically integrated stack (Blackwell, Vera Rubin CPUs, NVLink, Spectrum-X, InfiniBand, and CUDA software) creates a moat beyond silicon alone. NVIDIA projects $1 trillion in combined Blackwell and Rubin revenue from 2025 through calendar 2027, signaling confidence in sustained demand.

   AMD guided Q3 revenue to roughly $13 billion (+41% quarter over quarter) and expects Data Center to more than double year-over-year in 2027—success hinges on Helios yields and how quickly Anthropic's first gigawatt of MI450 GPUs converts to revenue. NVIDIA's ability to defend its 75% gross margin while China compute remains excluded from guidance will show whether geopolitical constraints limit upside. Vera Rubin, claimed to deliver up to 35× higher inference throughput than Blackwell, could extend NVIDIA's lead if it performs as promised.

2. 2

   AI startup makes 110-minute feature film for $2M in four weeks

   Higgsfield AI, a startup, produced The Cully Hill Boys, a 110-minute action-comedy featuring licensed celebrity likenesses, in just four weeks with a budget of $2 million—half spent on AI tokens. The film follows East London rappers who steal a boat for a music video and get caught between crime cartels. The film demonstrates AI video production can now scale to feature length, a shift from short clips. While the technology still shows visible flaws—unrealistic polish, occasionally obvious AI generation, and inconsistent character reactions—the human screenwriter's dialogue and story structure indicate AI tools may be moving toward democratized, cost-effective long-form production for creators.

   Higgsfield AI says it plans to provide scalable AI workflows for creators, signaling the company sees a market opportunity in bringing down the cost and time required for video production. The film's mixed execution—strong writing, visible generation artifacts—shows where the technology still needs refinement.

3. 3

   Developer builds pagedMark to strip AI watermarks from self-generated images

   A developer created pagedMark, a tool designed to remove AI provenance markers—both visible labels and invisible pixel-embedded watermarks—from images and videos generated by AI tools. The tool regenerates images to eliminate invisible marks while attempting to preserve visual similarity, and currently supports watermarks from ChatGPT, gpt-image API, Z-Image Turbo, and Nano Banana for images, plus visible marks and metadata from Sora, Veo, Seedance, Hailuo, and Kling for video. AI-generated content increasingly carries metadata and embedded signals (like SynthID watermarks) that identify its origin. For creators working with their own AI-generated material, these marks can complicate workflow—simple removal methods like screenshots don't eliminate pixel-level markers. pagedMark addresses this by regenerating affected content, though users should know the output may differ from the original (faces, text, and small details can shift).

   The tool has been tested on Apple Silicon Macs with 8 GB and 16 GB of memory, with memory-aware processing built in to handle resource constraints on lower-spec machines.

4. 4

   Can an LLM create a full-length film alone?

   A post on LessWrong poses the question of whether a large language model (LLM)—an AI system that understands and generates text—could independently produce a feature-length movie without human intervention. The question explores the boundaries of current AI capabilities in creative production, touching on whether text-based AI systems can handle complex, sustained creative tasks that require narrative coherence, structure, and sustained output across a feature film's runtime.

   The question itself remains open; the post frames it as an inquiry rather than a resolved finding, inviting discussion about the technical and creative limitations of LLMs in film production.

5. 5

   OpenAI's Senior Exodus Sparks Survival Questions

   OpenAI closed an $852 billion self-funded tender offer on August 10, 2026, then lost chief operating officer Brad Lightcap (August 11) and chief revenue officer Denise Dresser (August 13) within days. CNBC reported at least a dozen senior leaders have departed in 2026, including heads of ethics and safety. Ed Zitron's newsletter argues OpenAI's annualized run rate of $40 billion (roughly double 2025) masks decelerating growth, while planned compute spending of $750 billion through 2030 balloons obligations to above $800 billion—a gap he says requires raising $100 billion to $200 billion annually, which he calls "close to impossible." OpenAI president Greg Brockman countered that departures are unremarkable for a company in spotlight and that monthly run rate climbed 20% in July with 32% business-customer growth.

   Whether OpenAI files IPO paperwork at or near its $1 trillion target valuation (CEO Sam Altman has called anything lower a "non-starter"), whether Microsoft renegotiates commercial terms, and whether enterprise revenue continues outpacing consumer revenue—OpenAI's enterprise business topped consumer for the first time in the latest quarter.

6. 6

   Consolidated AI tools beat stitching together separate platforms for marketers

   A guide to AI image and video generators for marketers compares tools by their purpose rather than ranking them, finding that consolidated workspaces like Getimg.ai—which run image, video, music, speech and sound effect generation in one interface—reduce the friction of moving assets between separate applications. Getimg.ai's paid plans start at $8 a month on annual billing and grant commercial rights; the Entry tier (3,000 credits) converts to roughly 200 FLUX.2 images, or 100 Seedream images, or 10 Kling videos, or 60 Seedance videos, or 1,000 generated tracks. Marketers typically lose more time shuffling assets between tools than generating them. A consolidated workspace removes that export overhead and keeps brand references consistent across formats—critical for campaign production where revisions consume far more credits than initial generation. Getimg.ai publishes credit-to-output conversions explicitly (something almost no competitor does), letting teams forecast actual monthly costs rather than guessing from subscription price alone. For teams already invested in Adobe or Canva, staying within those stacks often wins on approval workflow alone, because assets never leave reviewers' existing systems.

   The guide identifies specialist tools for specific needs—Runway and Luma for cinematic video direction, Ideogram for text legibility inside frames, Creatify for performance marketing at volume. Free tiers across all platforms watermark output, cap resolution, and exclude commercial use, making them unsuitable for real campaigns; the practical entry point is a paid plan. Credits do not roll over month to month, so teams should size their plan to typical monthly volume (accounting for revisions) rather than peak output or promotional promises.