When Sora 2 was released (invite only), results were impressive. Investors were happy, users excited. Then, a strange thing happened. Sora began to buckle under load despite being invite only. In order to roll out Sora to all users, the technology looks to have been massively scaled back. Not only are users alleging OpenAI did this, the criticism is that they did so silently behind closed doors without elaborating. The results were clear. Overnight the quality of material generated by Sora 2 fell into the gutter like slop from a sausage machine. EOSHD takes a look at what has been going on… Sora 1 and the new model 2 are some of the most hyped AI tools of all time, driving enormous investor interest and at their best are capable of amazing visuals with very few imperfections. Prompts could be incredibly complex in Sora 2 at launch, involving several sections to inform the look. In OpenAI’s own example of the otherworldly dragon, you had a “First Read” Primary Target & Visuals to lay out the general scene, followed by a Format & Look using cinematography terms like 180 degrees shutter, large format sensor, crisp micro contrast and no gate weave. You could then go as far as to specify lenses, filters, grading and lighting. Sound is generated too. To round out the shot, the prompt can provide camera notes, location details and framing, as well as an optimised shot list to sequence it all. https://www.eoshd.com/wp-content/uploads/2025/12/dragon2.mp4 The result was a few...
Published By: EOSHD.com blog - Today