Midjourney 評測：AI 影片能否正確呈現文字？

You type a prompt into an AI image generator and get something technically correct but forgettable: flat lighting, a stock-photo look, and a composition you’d never pick yourself. Hiring a photographer or illustrator fixes that, but not when you need a hero image by this afternoon.

That’s the problem Midjourney has always been known for solving. Its images have a style that’s hard to mistake: cinematic light, rich texture, and compositions that look like someone made deliberate choices.

But a lot has changed since I first reviewed it. You no longer need Discord, V8.2 renders higher resolution 2K images, a new Edit Model handles reference images and inpainting, and Midjourney can now turn any image into a short video.

I tested Midjourney on three tasks: a product ad with text on it, a character I needed to keep consistent across several scenes, and a still image turned into a video clip.

In this Midjourney review, I’ll show you how it worked, where it impressed me, and where I think it still falls short. I’ll finish the article by comparing it to my top three alternatives: ImagineArt, ChatGPT Images 2.0, and GetIMG.

Verdict

Midjourney still makes some of the best looking AI images I’ve tested, and V8.2 finally got short text right and kept my character consistent across scenes. However, video generation is weak: it’s silent, it warped small text in the clips that I made, and HD burns through Fast hours quickly. Overall, Midjourney is worth paying for mainly if images are your priority.

Pros and Cons

• Great image quality with realistic lighting, texture, and composition.
• Native 2K image generation without needing to upscale.
• Much faster image generation than earlier versions.
• Up to four reference images to keep characters, products, and styles consistent.
• Built-in editing tools to fix or extend parts of an image.
• Style references and moodboards to keep your images consistent.
• Turns still images into 5–21-second videos with looping controls.
• Unlimited Relax-mode images on Standard and higher plans.
• Commercial use included with every plan.
• Works in your browser without needing Discord.

• No free plan or trial. You have to pay before you can generate images.
• Images are public by default. You need the $60/month Pro plan for Stealth Mode.
• Small text becomes distorted in videos, making them unsuitable for product ads where the label matters.
• Basic includes only 3.3 hours of Fast GPU time. HD video uses up those hours quickly.
• Video generation is limited to image-to-video at 720p.
• Prompting takes some practice, and several older features no longer work in V8.
• Thinking Mode and the Editor didn’t fix the backwards coat in my test.
• Midjourney is facing copyright lawsuits from Disney, Universal, and Warner Bros. Discovery.

What is Midjourney?

Midjourney is an independent AI research lab in San Francisco, founded by David Holz (co-founder of Leap Motion). Midjourney is self-funded and calls itself “community-funded,” and its main product is an image and video generator that sits alongside other AI graphic design tools.

You describe what you want in plain language, and Midjourney gives you a grid of images (four by default) in that style. From there, you can make variations, edit parts of an image, change the framing, upscale, or animate it.

What’s Changed

When I first tried Midjourney, you had to join its Discord server and type /imagine commands into busy public channels. Now most people use the web app, which works more like a regular creative tool: a prompt bar, a gallery of your work, folders, and an editor. Discord still works, but it’s no longer required.

The model has also changed a lot:

• V7 (April 2025) brought Draft Mode and Omni Reference.
• V8.1 (April 2026) added native 2K output and much faster rendering.
• V8.2 (default since July 2026) focuses on aesthetics and Personalization. Midjourney describes its look as “more creative, bold, sophisticated, edgy and fresh.”

V8.2 also introduced the Edit Model, which replaces three older tools (Omni Reference, Character Reference, and Retexture) with one system. You can attach up to four images and describe what you want changed, much like you would in a chat-based editor.

In October 2026, Midjourney started testing Thinking Mode. It looks at an image, finds where it missed your prompt (wrong objects, layout, anatomy, or text), and generates it again. It’s still in alpha.

It Makes Video Now

Midjourney’s video model animates an image you generated or uploaded. Clips start at 5 seconds and can be extended in 4-second steps up to 21 seconds. You can loop a clip, set an ending frame, and choose low or high motion.

Animations are a useful add-on, but it isn’t a full AI video generator: there’s no text-to-video, and resolution tops out at 720p.

What Midjourney Isn’t

Midjourney isn’t a dedicated design tool like Canva. It works best when you want a high-quality image and don’t mind making a few adjustments to get it right. It handles short text much better than before, but it’s still not great for designs with lots of text or when you need every detail to be just right.

It’s also worth knowing that Disney and Universal sued Midjourney in June 2025, and Warner Bros. Discovery followed that September. Meanwhile, Meta signed a deal in August 2025 to license Midjourney’s technology.

Who is Midjourney Best For?

Here’s who Midjourney is best for:

• Marketers and social media managers can create scroll-stopping campaign visuals and social media graphics in a consistent style, without booking a shoot.
• Concept artists and game developers can explore dozens of character, creature, and environment ideas, then use reference images to keep the strongest designs consistent. It pairs well with AI game generators for prototyping.
• Authors can mock up book covers and chapter illustrations and add basic title text.
• Bloggers can create hero images that don’t look like the same stock photos everyone else uses.
• Graphic designers can build moodboards and style references for client pitches, then hand the direction to a production team.
• E-commerce brands can create lifestyle and product-concept images, though they’ll still want real photos for exact packaging.
• Creators who want short video clips can animate a still into a looping background, intro, or social post.
• Hobbyists on the Standard plan can generate as many Relax mode images as they like for one flat price.

Midjourney Key Features

Here are the key features that stood out to me:

1. V8.2 image model: The latest version, designed for bold, eye-catching images. V8.1 also made standard jobs about four times faster.
2. Native 2K (HD) mode: Generates 2048px images directly with –hd. It costs about 1.3 minutes of GPU time per job, compared with 0.8 for SD.
3. Edit Model: Attach up to four reference images and describe the change. This one tool handles consistent characters, product placement, new angles, and restyling.
4. Editor (inpainting and outpainting): Paint over an area to change just that part, or extend the canvas past the original frame. Note that edits drop HD images back to SD until you upscale again.
5. Style references (–sref): Use a style code to copy a look and apply it to prompts. It’s one of the most useful ways to keep a series consistent.
6. Moodboards & Personalization: Rate images or upload a set of examples, and Midjourney builds a profile of your taste and applies it to future generations.
7. Image-to-video: Animates any image into 5-second clips that you can extend up to 21 seconds with loop, end-frame, and low/high motion controls. SD is 480p and HD is 720p.
8. Raw mode: Turns down Midjourney’s default styling so the image follows your prompt more literally. It’s useful for realistic product shots.
9. Thinking Mode (alpha): Checks a V8.2 image against your prompt and tries to fix anything that’s missing or wrong, like objects and text.
10. Shared folders: Invite collaborators to generate into a shared folder, or share it by link. It’s useful for small creative teams.

Hands-On Testing: What Midjourney Produced

I ran three tests based on the work people actually pay Midjourney for: a marketing image with text, a consistent character across scenes, and a short video. For each one, I focused on the result: how good it looked, how closely it followed my prompt, and how much I’d need to fix before using it.

I used V8.2 on the Standard plan ($30/month) with HD turned on, Raw mode off, and the other settings left at their defaults.

Test 1: A Product Ad With Text on the Label

Text inside images has always been Midjourney’s weak spot, so I wanted to see whether V8.2 had finally fixed it.

Here’s the prompt I used:

“A product photo of a matte black coffee bag on a sunlit wooden café counter. The label reads ‘NORTHBOUND ROAST’ in bold white letters, with ‘Single Origin · Ethiopia’ underneath in smaller type. Steam rising from a ceramic cup beside it, shallow depth of field, warm morning light, shot on a 50mm lens. –ar 4:5”

Midjourney returned four images in about 20 seconds, and every one of them got both lines of text right. “NORTHBOUND ROAST” and “Single Origin · Ethiopia” were spelled correctly on all four bags, with no extra letters, swapped characters, or melted type.

The rest of the prompt held up just as well. All four bags were matte black, all four sat on a sunlit wooden café counter, and the steam rising from the cup looked convincingly real. I didn’t spot any artifacts or distortions around the bag edges, the cup, or the counter.

My Take

I was genuinely impressed. I expected at least one garbled word and planned to add the label in a design tool afterward. Instead, I could have used any of the four images in an ad as-is without touching them.

Because nothing was wrong, I didn’t need to rerun anything. One generation, 20 seconds, four professional-looking images.

But the caveat is that I only tested two short lines of text. A brand name and a tagline are what most product shots need, but I wouldn’t assume the same accuracy for dense text like a menu or an infographic without proofreading it.

Test 2: Keeping One Character Consistent Across Three Scenes

Consistent characters matter for anyone making a comic, a children’s book, or a brand mascot. The first thing I did was create a character:

“A children’s book illustration of a small red fox wearing a yellow raincoat and green rain boots, standing in a puddle, soft watercolor style, white background.”

Midjourney gave me four options in about 15 seconds. I picked my favorite, attached it as a reference image in the Edit Model, and asked for three new scenes:

1. “The same fox reading a book under a mushroom during a rainstorm.”
2. “The same fox riding a bicycle through an autumn park, seen from the side.”
3. “The same fox asleep in a cozy bed at night, raincoat hanging on a hook.”

I ran each scene as its own generation, with the same original fox attached each time.

Every scene was usable, it was just a matter of preference. The fox kept the same face, ear and tail markings, yellow raincoat, and green boots for every generation, and the soft watercolor style held throughout. Even the side view on the bicycle (which I expected to be hardest) showed no wrong colors, no different snout, and no extra limbs.

The one problem was in the bedtime scene. In all four images, the raincoat was hung on the hook backwards by its hood, which isn’t how anyone hangs a coat.

I tried to fix it two ways.

First, I reran my favorite image with Thinking Mode. The rest of the scene was fine, but the coat got worse: the hook now went straight through the hood, which made the mistake more obvious.

For my second attempt, I opened the Editor, used Smart Select to select just the coat, erased it, and prompted: “a yellow raincoat hanging from the hook by its collar, hood falling down the back.”

That didn’t work either. The coat was still backwards, it had less detail than before, and the hood was bigger and gaping open. The watercolor style stayed consistent through every version, at least.

My Take

On the one hand, the character consistency impressed me. I’d use the reading and bicycle scenes in a picture book as they are.

But the bedtime scene is a different story. A backwards coat is a small detail, but it’s exactly what a child or parent would notice, and neither of Midjourney’s fixing tools could solve it. Thinking Mode is meant to catch where an image misses the prompt, yet it made this mistake more obvious, and the Editor made it worse.

To use that scene, I’d have to regenerate more grids, crop the hook out, or fix it in Photoshop.

Three scenes isn’t a whole book, but I’d trust the Edit Model with a character. I’d just budget time to check every image for details that don’t make sense.

Test 3: Turning a Still Image Into a Video

For my final test, I took my favorite coffee bag image from Test 1 and animated it with this prompt:

“Steam slowly curls up from the cup, the camera pushes in gently toward the bag, morning light flickers through the window.”

I generated a batch of four clips in HD (720p) with low motion and no end frame or loop, then extended my favorite clip with the same settings.

The four clips took about 2 minutes and used 26 Fast minutes, which Midjourney warned me about in a pop-up before it started. The motion was smooth and realistic, and the big “NORTHBOUND ROAST” label stayed sharp. But the smaller “Single Origin · Ethiopia” line warped in all four clips.

Extending my favorite clip cost another 25 Fast minutes and gave me four more versions. They carried on smoothly from the original, but the same small text was still distorted. Also, all four videos zoomed in on the steam, which made the steam look more obviously like blobs as opposed to relistic wisps. None of the clips had audio.

My Take

The motion quality was better than I expected from an image tool, but warped text rules these clips out for a product advertisement where the label is the whole point. A shot without small text would work much better.

The cost is the bigger issue. One 9-second HD clip and its extension used about 51 Fast minutes, which was more than three times what both of my image tests used together. Dedicated video generators also give you text-to-video, audio, and longer clips, so I’d treat Midjourney’s video as a bonus, not a reason to subscribe.

Results & Output Quality

• Image quality: Midjourney’s strongest point was how finished its images looked. The coffee-bag shots had believable morning light, steam that looked real (for the most part), and great composition. Meanwhile, the fox kept a soft watercolor style in every scene.
• Prompt accuracy: V8.2 followed my prompts closely. Every bag was matte black on a sunlit wooden counter, and every scene I described matched what I asked for. The one miss was a small detail: the fox’s raincoat hung backwards on its hook.
• Text: All four images in Test 1 spelled both lines of the label correctly on the first try, so I never needed Thinking Mode. But when it came to turning those images into videos, that’s where the small text became distorted and unreadable.
• Consistency: My fox character stayed consistent across all of the scenes I generated. However, the fox’s coat was hanging backwards on the hook, and neither Thinking Mode nor the Editor could fix it.
• Video: The video’s motion was smooth, but the small text warped in all eight clips, making none of them usable for a product ad.
• Speed: Image grids took about 15–20 seconds in HD. A batch of four HD videos took about 2 minutes. That’s some of the fastest image and video generation I’ve seen, especially for this level of quality.
• Editing required: My Test 1 images and two of my three fox scenes needed no edits. The bedtime scene would need more editing attempts, a crop, or Photoshop.
• Value: All three tests used about 1 hour and 6 minutes of my 15 Fast hours. Images took about 15 minutes of that and HD video about 51. On Standard, images are excellent value, but HD video eats into Fast hours quickly.

How to Use Midjourney

Here’s how to get good results on Midjourney:

1. There’s no free trial, so pick a plan and sign in. If you plan to generate a lot, Standard’s unlimited Relax mode is usually better value than Basic.
2. Set up Personalization or a moodboard first. Uploading 10–20 examples of a style you like gives every future prompt a head start.
3. When writing a prompt, describe the subject, setting, lighting, lens or medium, and mood. Add parameters at the end, like –ar 16:9 for widescreen or –raw for a more literal result.
4. Pick the best image and refine it. Use Vary for close alternatives, the Editor to fix one area, or the Edit Model with reference images to keep a character or product consistent.
5. Upscale or animate. Export the final image in HD, or choose Animate to turn it into a short clip.

If you want to get better at prompting in general, these prompt engineering courses are a good place to start.

Top 3 Midjourney Alternatives

Here are the best Midjourney alternatives I’ve tried that I’d recommend.

ImagineArt

ImagineArt is the best fit if you want images, video, and audio in one subscription and don’t want to be tied to a single model. Midjourney only offers its own models. ImagineArt gives you access to third-party image and video models like Nano Banana, GPT Image, and Kling, plus voiceovers and music.

That’s the main trade-off. Midjourney has a distinctive default aesthetic, while ImagineArt lets you choose a model for each job. This gives ImagineArt more flexibility for tasks like rendering readable text or generating videos. However, results can vary more between models.

ImagineArt is also cheaper to start. It offers a free plan with 100 credits that refresh daily, whereas Midjourney doesn’t offer a free plan or trial. It also has iOS and Android apps.

Choose Midjourney for the best-looking images straight out of the box. Otherwise, choose ImagineArt if you want an all-in-one creative toolkit with more models and video tools.

Read my ImagineArt review or visit ImagineArt!

ChatGPT Images 2.0

 The second Midjourney alternative I’d recommend is ChatGPT Images 2.0. It’s better at getting the details right than the overall look.

In my review, it got text-heavy images, infographics, and detailed instructions right far more often than most image generators. Midjourney V8.2 has closed the gap on short text (it spelled my product label correctly in all four images), but ChatGPT is still the safer bet for text-heavy designs.

The workflow is also different. Instead of writing a prompt and choosing from a grid, you talk to ChatGPT (“make the cup blue”) and it edits the image. When I edited my images in ChatGPT, it made the changes I asked for.

But where Midjourney still wins is style and speed. ChatGPT’s images often look less cinematic, and its Thinking mode took about three and a half minutes for an eight-image set in my testing.

Choose ChatGPT Images 2.0 for accurate text, detailed designs, and edits that follow your instructions. Otherwise, choose Midjourney for better image quality and faster results.

Read my ChatGPT Images 2.0 review or visit ChatGPT!

GetIMG

My final recommendation (GetIMG) gives you access to many of the leading AI models in one place. On the image side, that includes Seedream 5, FLUX.2, Nano Banana, and GPT Image. For video, it offers Kling, Seedance, Veo, and Wan.

That makes it the opposite of Midjourney, which has its own models and a distinctive default aesthetic. With GetIMG, you can run the same prompt through several models and keep the best result, which helps when one handles text better and another handles photorealism.

Pricing works on credits, with Core unlocking every model. Choose GetIMG if you want to experiment with different models instead of sticking to one signature aesthetic.

Read my GetIMG review or visit GetIMG!

Midjourney Review: The Right Tool For You?

What I liked most about Midjourney is that its images look carefully designed. My coffee-bag ad was ready to use in about 20 seconds, with both lines of label text spelled correctly. My watercolor fox also looked like the same character across all three scenes.

Where it fell short was video. The motion looked smooth, but small label text warped in every clip, the extensions made the steam obvious they looked more like blobs, and one HD clip plus its extension used about 51 Fast minutes. Paying $60/month just to keep your images private also feels steep for solo creators.

And then there was Thinking Mode. It’s meant to fix places where an image misses your prompt, but on my backwards-hung raincoat, it made the mistake more obvious. The Editor didn’t do any better, so Midjourney’s editing tools didn’t live up to my expectations.

I’d recommend Midjourney to marketers, designers, concept artists, and authors who care most about how an image looks. For images, the Standard plan is the sweet spot: my Test 1 and Test 2 results needed little to no editing, and unlimited Relax mode means you can keep generating.

But if video, text-heavy designs, or a free plan matter more to you, one of these alternatives might be a better fit:

• ImagineArt is best for creators who want images, video, voice, and music from many models in one cheaper subscription.
• ChatGPT Images 2.0 is best for accurate text, infographics, and editing images through conversation.
• GetIMG is best for comparing many top image and video models side by side.

Thanks for reading my Midjourney review! I hope you found it helpful. If you want to try Midjourney for yourself, sign up here and see what it can build for you.

Frequently Asked Questions

Is Midjourney free?

No. Midjourney has no free plan or free trial. You need a paid plan before you can generate anything. 

Can I use Midjourney images commercially?

Yes. You own the images and videos you create on any paid plan, even after you cancel. However, companies earning over $1 million a year must use the Pro or Mega plan. Midjourney doesn’t give copyright advice, so check the rules for your country before using images as part of your business tools. If you need print-size files, a dedicated AI image upscaler can take its 2K output further.

Do you still need Discord to use Midjourney?

No. You can create, edit, organize, and animate images entirely on the Midjourney website. Discord still works if you prefer it, and Midjourney’s support and community are still largely based there.

Is Midjourney better than ChatGPT for images?

It depends on the job. Midjourney usually produces more stylized images, while ChatGPT Images 2.0 is better at designs with lots of text and edits you describe. However, Midjourney V8.2 spelled a short product label correctly in all four images in my testing.