AI Trends

Kling Image 4.0 Hands-On: From Image Generation to Autonomous Visual Design

LeiphoneUpdated Nov 21 11:58

AI summary

Leiphone tested Kling Image 4.0 and found that its capabilities extend beyond portrait and poster generation to long-form e-commerce images and continuous storyboard design, with support for editing text in specific areas. The new model’s native resolution has been increased to 8K; each image can contain up to 2,000 Chinese characters, it can generate up to 10 coherent images at once, and it includes more than 70 preset fonts. During testing, some small explanatory text lacked clarity, while origin and product information added by the model was not always accurate.

Why read this

SelectedAI score 79

The hands-on evaluation covers posters, long-form e-commerce images, and continuous storyboards, helping assess the model’s capabilities and limitations as it moves from generating individual images to designing complete visual solutions.

Topics

Tags

Body · AI translation

Over the past few years, AI image-generation models have continually reshaped perceptions of realism, resolution, and lighting. But when these images are actually used in commercial settings such as posters and advertisements, questions remain over whether text can be rendered accurately and whether localized edits can preserve the original design.

These seemingly minor requirements remain difficult challenges for image models to this day. (Leifeng.com)

In March this year, Microsoft Research released BizGenEval, a benchmark for commercial visual generation that tests leading models with 400 prompts and 8,000 verification questions. The results showed that models can already generate visually appealing posters and web pages, but still have clear shortcomings in areas such as precise layouts and correctly matching attributes to multiple objects.

It is at precisely this stage that Kling, best known for its video-generation capabilities, launched Kling Image 4.0 today.

Generated with Kling Image 4.0

According to the official introduction, the new model increases native resolution to 8K, supports up to 2,000 Chinese characters in a single image, and can generate as many as 10 coherent images at once. More notably, it uses a unified autoregressive architecture for understanding and generation, attempting to let the model handle image understanding and generation within the same context.

This time, Kling clearly wants to do more than simply make images prettier; it also hopes to enable the model to handle more complex creative requirements.

We began with the most intuitive test—portrait photography—to explore Kling Image 4.0’s capabilities. From taking a good photo and designing a poster to organizing an entire suite of visual content, just how much work can Kling Image 4.0 do for creators? 01

From Portraits to Creative Posters

How powerful is Kling Image 4.0 at generating images?

What surprised us most during our experience with Kling Image 4.0 was the lifelike quality of its generated images. We tried creating several portraits, and whether in the subjects’ skin texture and details or the interplay of outdoor light and shadow, the results were already remarkably close to real photography.

This degree of realism has begun to affect the photography industry itself. In April this year, Vogue interviewed a number of commercial photographers and found that some clients had started using AI-generated images as references for shoots, asking photographers to recreate those scenes in the real world.

Portraits, however, primarily test a model’s ability to reproduce people and realistic settings. Creating a poster genuinely intended for public distribution is considerably more complicated, especially once Chinese text is added.

Recently, an exchange between guests Xu Fei and Li Na on a Mango TV variety show unexpectedly turned the phrase “draw a circle on the ground as a prison” into a new internet catchphrase. We decided to let Kling Image 4.0 join in on the trend.

Compared with an ordinary text-based poster, “draw a circle on the ground as a prison” is actually a highly suitable theme for visual design. It has both clearly defined textual content and imagery that can be expressed visually. The final work strongly resembles a promotional poster for a Mango TV variety show: from the semicircular seascape composition and cool, detached color palette to the typographic layout and artistic lettering, it nearly reaches the standard of a publication-ready poster.

One change deserves special mention: Kling Image 4.0’s image editor can directly select and modify text already present in an image, as well as add new text to an existing composition. Version 3.0 already offered localized re-editing, but 4.0 provides finer-grained control. We tried changing the poster’s wording and adding a line of secondary copy. Only the requested text changed in the original poster; the other image layers were not redrawn.

Font control is another area where 4.0 has pulled ahead of 3.0. In version 3.0, creative lettering mainly relied on prompts describing the desired style. It worked, but offered limited control. Version 4.0 turns this into something more like a practical tool: it includes more than 70 preset fonts that can be selected directly and also supports uploaded font-reference images, allowing the model to generate text that follows the letterforms and stroke style of the reference.

For example, in the summer camping poster we created below, the artistic lettering blends very naturally into the scene. From the setting and color palette to the title design, it has the polished feel of a nearly complete final product and could serve as an event promotional graphic after only minor checks and adjustments. The lettering itself can become part of the composition, interacting with clouds, waves, and other visual elements. For the model, this requires simultaneously managing the relationships among textual content, spatial positioning, and visual style.

In addition to creative posters with relatively little text, Kling Image 4.0 also claims to support up to 2,000 Chinese characters in a single image. Short titles place greater demands on typography and creativity, while longer passages are more challenging in terms of textual accuracy, information hierarchy, and layout. In the product detail pages that follow, we will continue examining how it handles more complex combinations of text and imagery. 02

A Product Launch for a Five-Yuan Roasted Sweet Potato

If artistic-lettering posters mainly test a model’s aesthetics and creativity, product detail pages involve far more complex tasks. Product photography, how content is presented on a detail page, and the sequencing and arrangement of text and images are all real-world commercial design scenarios.

This is also one of the capabilities enhanced in Kling Image 4.0. The new model supports more complex mixed layouts of text and images, can accommodate up to 2,000 Chinese characters in a single image, and can create long-form e-commerce detail graphics with one click.

We chose an utterly ordinary product: a roasted sweet potato. We wanted Kling Image 4.0 to use this photo as the basis for designing e-commerce promotional materials for a roasted sweet potato, including appealing product photography as well as key selling points, consumption scenarios, and essential product information.

Prompt: Based on the real photo of a roasted sweet potato I uploaded, please design a complete long-form e-commerce detail page for this ordinary roasted sweet potato, priced at five yuan, with the look and feel of a product launch by a leading global technology brand such as Apple.

The model’s result really did look like something from an established food brand.

It designed a brand identity called “ROOTS Root Vegetable Series” for the product, using green and off-white as the main visual elements and “Reddish-Brown and Crisp, Golden and Molten” as the headline. The previously ordinary roasted sweet potato was placed in a carefully arranged product photography setting, with its warm golden flesh and dark brown skin becoming the visual focus.

More interestingly, the model did not cram all the content into the main product image. Instead, it extended the design into several different types of pages: one showing a serving scenario with sliced pieces arranged on a plate, another describing the product’s origins under the heading “The Terroir of Sunlit Hills,” and a third organizing different serving states alongside storage and reheating instructions into an illustrated guide.

Looking back carefully, the prompt was only one simple sentence and provided no complete product information. Yet the model proactively supplied the brand name, advertising copy, and other information it had been asked to design, organizing everything into a stylistically consistent set of visuals.

This was also the most surprising part of our test.

In the past, using AI to generate images often meant first deciding exactly what the image should look like, then describing it to the model in as much detail as possible—such as the product’s position, the background color, and the text layout. Producing an entire set of promotional materials also required designing prompts image by image and repeatedly adjusting the style. This time, however, we provided only one photo and a relatively broad objective. Kling Image 4.0 was able to independently develop a substantial portion of the more detailed visual concept.

For small merchants without professional design teams, or operations staff who need to produce marketing materials quickly, this capability has clear and immediate value. Previously, they might have needed to prepare product photos, copy, and layouts separately. Now they can first generate an entire visual concept with a single natural-language sentence and then adjust it according to their actual needs.

Of course, there are still areas where the generated results could be improved. For example, some of the small explanatory text is less clear than the large headlines, and the place-of-origin and product information added by the model may not be accurate. Overall, however, Kling Image 4.0 has already demonstrated a fairly comprehensive ability to organize commercial text and imagery. Rather than merely generating an attractive product image, it can go on to handle brand styling, advertising copy, scenario displays, and page layouts, rapidly turning a simple idea into a concrete, visible design concept. 03

From One Image to a Series

In recent years, image models have begun exploring the generation of image series. Films and comics both require continuous storyboards, while brand advertising often requires a complete set of visual materials. Questions such as how the same character can change across different images while remaining consistent, and how shots should transition from one to the next, cannot be solved simply by improving the quality of individual images.

Kling Image 3.0 already supported continuous image series. With version 4.0, Kling has further strengthened consistency in subjects and scenes across multiple images. According to the official introduction, the new model can generate up to ten interrelated images at once.

We selected a still from the film Portrait of a Lady on Fire as a reference and asked the model to independently generate a set of cinematic storyboards centered on a story in which two women reunite by the sea and burn an unsent letter.

Still from Portrait of a Lady on Fire

Judging from the results, the model is already capable of constructing a fairly complete visual narrative.

Selections from the 10 generated cinematic storyboard images

The most striking feature of this image series is that the model proactively used different shot scales—including long shots, medium shots, and close-ups—to progressively develop the visual narrative. In particular, the transition from handing over the letter, to a close-up of the envelope, and then to the characters’ facial expressions already demonstrates a fairly clear awareness of shot organization.

The story begins with a long shot of the two figures on the coast, then cuts to a two-shot in which one person hands an envelope to the other. The camera immediately pushes in for a close-up of the envelope. In the fourth image, it cuts to a close-up of a character’s face as she looks down at the letter in her hands. The bonfire from the reference image appears in the fifth frame, and the sequence ends with a long shot of the two women standing on opposite sides of the firelight.

The generated images also maintain a relatively consistent visual style. The dark clothing and firelit atmosphere of the reference still carry through, while the overcast coast, deep-blue twilight, rocks, and warm golden firelight recur throughout the series.

Of course, close comparison still reveals a few minor issues. For example, the characters’ faces vary subtly between shots, and some transitions between actions are not entirely natural. Nevertheless, for this test, which used only a story synopsis as input, the model produced a set of draft storyboards with a clear narrative sequence and a unified visual style.

(A set of animation storyboards we asked Kling Image 4.0 to generate)

For preproduction in film and television, comic storyboarding, or brand advertising planning, this capability opens up considerable possibilities. Creators can first provide a story, character settings, and visual references, have the model generate an initial set of images, and then select suitable shots for further refinement.

In the real-world film and television industry, AI image-generation tools like this are also making their way into preproduction. This June, American director Martin Scorsese, who has been making films for decades, told The New York Times that he had been drawing storyboards for about 70 years. One challenge he has always faced is how to accurately convey the scenes he imagines to cinematographers, production designers, and other crew members. Now, he can first use an image model to render those scenes, enabling the team to understand and discuss them more quickly.

This is very similar to the scenario in our test. Using only a story synopsis and one reference image, Kling Image 4.0 completed the first round of storyboard ideation. It serves more as a starting point for subsequent creative discussion. This is probably what Scorsese meant by “enabling the team to understand and discuss them more quickly”: the model first turns the images in one’s mind into something tangible that can be pointed to, while the actual judgment and revisions remain in the creators’ hands. 04

After Generation

These are actually two different capabilities: one is understanding the user’s intent and filling in parts they have not yet fully thought through; the other is accurately following instructions while interfering as little as possible with an established design.

Kling Image 4.0’s technical upgrades can be understood in this context. According to Kling, the new model uses a unified autoregressive architecture for understanding and generation, and handles more complex creative requirements through an instruction-processing capacity of up to 15K tokens, structured visual grammar, and agentic task-planning capabilities. Ultimately, these technical terms address whether the model can understand a task, properly organize its details, and preserve the parts the user is already satisfied with when revisions are needed.

These changes are not limited to Kling. This November, Adobe launched new creative agents and design tools in an effort to bring ideation, generation, and revision into a single working environment. Kling’s Video 4.0, released in October, also placed greater emphasis on multiple keyframes, reference materials, and camera control. Its latest product now further expands its visual design and editing capabilities.

For Kling, there are also more practical commercial considerations behind this. Kuaishou’s financial report shows that in the second quarter of 2026, Kling AI generated more than RMB 850 million in revenue, representing year-on-year growth of over 200%. Video generation has already brought it substantial revenue, but as commercialization reaches this stage, a new question has emerged: how can users be encouraged to move from occasionally generating a video to continuously using Kling throughout their creative work?

The image model upgrade creates more possibilities in this regard. Kling is competing for users’ ongoing creative needs on the platform. A stunning image or video determines why users open Kling for the first time; whether it can support the entire process from inspiration to finished product may determine why they keep using it.

Click a keyword to view the corresponding hands-on test

English text is being prepared.

English text is being prepared.

GLM 5.1 vs DeepSeek V4 Pro vs Kimi K2.6 vs Qwen 3.7 Max vs MiniMax M3

English text is being prepared.

ByteDance Doubao 2.1 Pro Seedance 2.5 vs Kling 3.0 Minimax H3 vs Seedance 2.0 Fast WorkBuddy vs Doubao Work vs Qwen Office More Reviews

Click to view 30 more reviews

Source

Leiphone