DALL·E 2 Competiton - Everything We Know about the Midjourney SEQUEL & UPDATE

TL;DR

Midjourney’s update adds a new upscaler and introduces Stylize and Quality controls that directly change artistic look, generation time, and cost.

Briefing Cornell Notes

Briefing

Midjourney’s latest Discord update adds a new upscaler plus two new control knobs—Stylize and Quality—that let users trade cost and speed for different kinds of image refinement. The practical takeaway is that Midjourney can now generate faster, cheaper “draft” results (including a low-quality mode described as four times faster) or spend more time per image for higher-detail outputs, with Stylize controlling how artistic or “hands-off” the generations look. Early comparisons in the transcript suggest the new Quality settings boost detail and sometimes creativity, while coherence improvements appear more limited than what users associate with DALL·E 2’s more consistently “real-photo-like” structure.

That matters because the update arrives amid growing complaints about DALL·E 2’s recent accuracy and consistency. After DALL·E 2 opened a beta that allows up to a million people to join via a paid prompt-based system, some users reported that image quality has slipped—showing up as less accurate generations, more “random stock image” outcomes, and failures to follow prompts. One example described an intended award-winning impressionist still life (mixing bowl, pickled cucumbers, socks, sunset) that instead produced unrelated scenery elements. While some of these issues are framed as possible bugs—such as caching/routing problems—or side effects of adding a diversity filter, the broader point is that reliability has become a live concern for DALL·E 2 users.

Against that backdrop, Midjourney’s “big” next step is rumored to be a sequel-scale model rather than a simple tweak. Community chatter points to a “4 billion model,” with claims that it could be trained on roughly 4 billion images and have around 2 billion parameters—positioned as more efficient than DALL·E 2, with curated outputs aimed at better results. The transcript emphasizes that these details are unconfirmed, but it also frames the upcoming change as potentially recognizable: Midjourney’s current aesthetic is described as distinctive, and a new model may look and behave differently enough to feel like a new generation.

The transcript also notes a platform shift: the current Midjourney version may eventually be invited into other servers, expanding access beyond a single Midjourney Discord space. Meanwhile, the current system is already undergoing alpha testing of a separate algorithm update, with warnings that some things may break and that further changes could land within one to three weeks.

Hands-on tests in the transcript compare chicken images across settings and models. With the new Midjourney controls, turning Stylize off reduces the “artistic” look but still leaves some stylization, while higher Quality settings increase detail at the cost of time and compute. DALL·E 2 is contrasted as producing more hyper-realistic chicken results by default, and it can also generate stylized chickens when prompted for stylization. The transcript ends by proposing a follow-up experiment: using Midjourney’s output and then leveraging DALL·E 2 to enhance coherence, potentially combining Midjourney’s creativity with DALL·E 2’s realism.

Cornell Notes

Midjourney’s update introduces a new upscaler and two new generation controls: Stylize and Quality. Stylize adjusts how artistic the output is, while Quality changes how long the model spends generating—offering faster/cheaper “draft” modes up to slower/more expensive settings that can increase detail and creativity. The transcript contrasts this with DALL·E 2, where some users report declining prompt accuracy and coherence after recent beta pricing changes, including examples where outputs miss the intended subject matter. Looking ahead, community rumors describe a larger “4 billion” Midjourney sequel model trained on billions of images, potentially improving curated results and coherence, though nothing is confirmed. The update also hints at broader Midjourney availability beyond its current Discord server.

What new controls did Midjourney add, and how do they affect the tradeoff between speed, cost, and output style?

Midjourney added Stylize and Quality settings. Stylize controls how stylized or artistic the generations are, ranging from turning stylization off to higher “hands off the wheels” options. Quality controls generation time and cost: a low-quality mode is described as four times faster and cheaper (rough drafts), the default is around “1,” and higher settings slow generation and increase price. Quality up to 5 is described as experimental, taking about five minutes per image and potentially producing more creative or detailed results, though it can also be worse.

Why are DALL·E 2 accuracy complaints showing up more prominently in the transcript’s discussion?

After DALL·E 2’s paid beta expansion, some users reported that image quality and prompt accuracy have declined. The transcript cites examples where generations became less accurate—producing unrelated elements instead of the requested scene—and suggests possible causes such as bugs (caching/routing issues) or side effects from adding a diversity filter. The key claim is that reliability appears to have worsened for some users, even if it’s debated whether the issue is real or simply noticed more due to increased usage.

How does the transcript characterize the difference between Midjourney’s creativity and DALL·E 2’s coherence?

Midjourney is described as strong on creativity, with Quality settings sometimes improving detail and coherence only slightly. DALL·E 2 is associated with stronger coherence—outputs that more consistently resemble structured, realistic scenes. In the chicken comparisons, DALL·E 2 tends to produce hyper-realistic chicken images by default, while Midjourney’s outputs remain more stylized and “opinionated,” even when stylization is turned off.

What is known versus rumored about the next major Midjourney model?

Known: Midjourney is already rolling out an alpha testing phase for a separate algorithm update, including a new upscaler and the new Stylize/Quality controls. Rumored: a larger “4 billion model” is widely discussed in the community, with claims about training on roughly 4 billion images and parameter count around 2 billion. The transcript stresses these details are unconfirmed and could change, including timing (rumored to be within the next few months).

What experiment does the transcript propose to combine Midjourney and DALL·E 2?

After generating a chicken with Midjourney’s new settings, the transcript proposes uploading that image and using DALL·E 2’s edit/image workflow to remove disliked parts (notably the face) and then regenerate with a prompt like “stylized photo of a chicken.” The goal is to keep Midjourney’s creative look while using DALL·E 2 to add more realism and coherence.

Review Questions

How do Stylize and Quality differ in what they control, and what are the described cost/speed implications for each?
What kinds of DALL·E 2 failures are cited as evidence of reduced accuracy, and what explanations are offered for them?
What specific rumored training/parameter figures are mentioned for the next Midjourney model, and why does the transcript treat them cautiously?

Key Points

1
Midjourney’s update adds a new upscaler and introduces Stylize and Quality controls that directly change artistic look, generation time, and cost.
2
Stylize ranges from turning stylization off to higher settings that can produce more extreme, “hands-off” artistic results.
3
Quality settings are described as a speed/cost ladder, from faster draft-like outputs to slower, more expensive experimental modes up to Quality 5.
4
Some users report DALL·E 2 prompt-following and accuracy issues after its paid beta expansion, with examples where outputs miss the intended subject matter.
5
The transcript frames Midjourney as stronger on creativity while DALL·E 2 is associated with more consistent coherence and realism.
6
Community rumors describe a larger Midjourney “4 billion” sequel model with training on billions of images and a parameter count around 2 billion, but nothing is confirmed.
7
A future workflow idea combines Midjourney’s output with DALL·E 2 editing to improve coherence by fixing parts like the face.

Highlights

Midjourney’s new Quality setting is explicitly positioned as a compute dial: lower Quality is faster/cheaper, while Quality 5 can take about five minutes per image and may be more detailed—or worse.

DALL·E 2 users in the transcript describe a drop in prompt accuracy, including cases where requested scenes devolve into unrelated elements.

The next Midjourney step is framed as potentially a sequel-scale model (“4 billion”) rather than a minor update, with rumored training and parameter figures.

The transcript proposes a hybrid approach: generate creatively in Midjourney, then use DALL·E 2 editing to correct coherence and realism.

Topics

Midjourney Update
DALL·E 2 Accuracy
Stylize and Quality
Upscaling
Model Sequel Rumors